Pedestrian Detection

This paper surveys real-time object detection literature critically and analytically, focusing particularly on pedestrian detection for safe autonomous vehicles. It addresses the challenges in the domain, some of the sources of which are variations in age, gender, clothing, lighting, backgrounds, and occlusion. The paper reviews object detection algorithms after providing an overview of deep learning basics and main architectures of neural networks, followed by discussion on existing algorithms along with their strengths, weaknesses, and future research directions. There is a need for pedestrian detection datasets with further complex annotations and multi-source integration, which captures interactions between pedestrians and their surroundings. Incorporating advanced sensors, including LiDAR, infrared, and depth sensors, as the foremost means to enhance the detection capabilities in more adverse conditions, such as low-light situations and occlusion. However, architectures such as YOLO, SSD, and Faster R-CNN, which have led to current improvements in performance, still allow room for improving pedestrian detection accuracy. By filling in these insights and proposed solutions, the paper focus on the development of pedestrian detection technology, how it can be brought into a safer, reliable, real-world applicability towards the system of autonomous driving. All of these results point to continued innovation towards deep learning, multi-sensor integration, and developing datasets to achieve optimal performance levels in real world conditions for autonomous driving systems.

Vehicle Detection algorithms run on deep neural networks. But one problem arises, when the vehicle scale keeps on changing then we may get false detection or even sometimes no detection at all, especially when the object size is tiny. Then algorithms like CNN, fast-RCNN, and faster-RCNN have a high probability of missed detection. To tackle this situation YOLOv3 algorithm is being used. In the codec module, a multi-level feature pyramid is added to resolve multi-scale vehicle detection problems. The experiment was carried out with the KITTI dataset and it showed high accuracy in several environments including tiny vehicle objects. YOLOv3 was able to meet the application demand, especially in traffic surveillance Systems.

Despite the rapid development of technologies for the detection of pedestrians in traffic, there are still significant problems with automated vehicle management systems. Often these problems are caused by different external conditions, weather disasters, different light levels, day and night, and more. All these factors can affect the incorrect assessment of the system and the performed detection of the pedestrian object. Pedestrian fatalities have risen recently, even as vehicles are equipped with sophisticated avoidance technology. Vehicle-based sensors can fail to identify pedestrians especially when those pedestrians are small, too far or too close to the vehicle, or partially occluded by nearby objects. Pedestrians have variable physical characteristics and appear in a variety of environments with different background features. The paper analyzes the possibilities and problems that arise in the detection and recognition of pedestrians. Pedestrian safety is one of the important...

Achieving pedestrian detection by means of computer vision is not a new topic in the field of computer vision research; however it is still being pursued with renewed interest because of the huge scope for performance improvement in the existing systems. Generally, the task of pedestrian detection (PD) involves stages such as pre-processing, ROI selection, feature extraction, classification, verification/refinement and tracking. Of all the steps involved in the PD framework, the paper presents the work done towards implementing the feature extraction and classification stages in particular. It is of paramount importance that the extracted features from the image should be robust and distinct enough to help the classifier distinguish between a pedestrian and a non-pedestrian, while a good classification algorithm would go a long way in precisely identifying a pedestrian as well as in simplifying the verification stage of the PD framework. The presented work focuses on the implementation of the Histogram of Oriented Gradients (HOG) features with modified parameters that can represent accurate intrinsic information of the image. Classification is achieved using Support Vector Machine (SVM). However instead of employing a readily available SVM library, the linear SVM implemented uses the Sequential Minimal Optimization (SMO) algorithm. The results observed by this HOG-SVM combination show promise to be the best feature extraction cum classification module for a full-fledged PD system.

As indicated by the Transportation Research and Industry Prevention Programme (TRIPP)'s Road Safety in India Report-2020, 33% of the accidents victims (deaths) are pedestrians. Heavy vehicles as well as cars are not able track pedestrian's movements on time. Most of the Children met with the accidents due to vehicle reversing. This problem motivates to track pedestrian through rear-view in heavy vehicles as well as for cars. Certain machine learning and deep learning approaches will best adapt to coping with the particular problems of rear-view pedestrian detection. In this work a literature survey of pedestrian detection and tracking research methodology and their constraints are discussed briefly. Most of the camera applications mainly concentrate on picture visibility and tracking. If the pedestrian detection application makes as inbuilt technique, then automatically so many accidents especially of children can be avoided. This pedestrian application mainly used to track the pedestrian movements while he or she is moving on heavy traffic roads and highways or while taking vehicle reverse by using cameras which were fixed on vehicles and make alerts. Such that camera can get more extracted features and helps the future applications. In this research paper a brief literature review is placed according to various researchers along with their techniques. And also compare the performance measures such as accuracy, sensitivity, false alarm rate and detection rate. These experimental results are out performance the methodology and differentiated with present technology.

With the decreasing cost of traffic cameras and rapid advancement in computer vision and artificial intelligence, developing robust traffic surveillance systems has become more feasible and practical. These systems can easily outperform traditional human monitoring systems, as they can collect and analyze traffic data coming from multiple cameras efficiently. A good understanding of this data allows the detection easily road anomalies in real time and in an autonomous way. Therefore, an intelligent traffic system typically consists of three components: object detection, object tracking, and behavior analysis components. In this paper, we present a review of some of the well-known object detection techniques used in traffic video surveillance. The review begins with a brief introduction to the history of object detection and the evolution of its techniques. Then we review separately the two main approaches of detection, which are traditional and deep learning approaches of detection. Finally, an experimental analysis has been conducted to evaluate and compare the performance of some of the recent relevant detection methods in terms of speed and precision, in detecting vehicles in a traffic scenario.

Traffic Sign Detection (TSD) is a complex and fundamental task for developing autonomous vehicles; it is one of the most critical visual perception problems since failing in this task may cause accidents. This task is fundamental in decision-making and involves different internal conditions such as the internal processing system or external conditions such as weather, illumination, and complex backgrounds. At present, several works are focused on the development of algorithms based on deep learning; however, there is no information on a methodology based on descriptive statistical analysis with results from a solid experimental framework, which helps to make decisions to choose the appropriate algorithms and hardware. This work intends to cover that gap. We have implemented some combinations of deep learning models (MobileNet v1 and ResNet50 v1) in a combination of the Single Shot Multibox Detector (SSD) algorithm and the Feature Pyramid Network (FPN) component for TSD in a standardized dataset (LISA), and we have tested it on different hardware architectures (CPU, GPU, TPU, and Embedded System). We propose a methodology and the evaluation method to measure two types of performance. The results show that the use of TPU allows achieving a processing training time 16.3 times faster than GPU and better results in terms of precision detection for one combination.

This paper proposes a feature-based technique to detect pedestrians and recognize vehicles within thermal images that have been captured during nighttime. The proposed technique applies the support vector machine (SVM) classifier on CENsus Transformed histogRam Oriented Gradient (CENTROG) features in order to classify and detect humans and/or vehicles. Although thermal images suffer from low image resolution, lack of colour and poor texture information, they offer the advantage of being unaffected by high intensity light sources such as vehicle headlights which tend to render normal images unsuitable for nighttime image capturing and subsequent analysis. Since contour is the most distinctive feature within thermal images, CENTROG is used to capture this feature information and is used within the experiments. The experimental results so obtained were compared with those obtained by employing the CENsus TRansformed hISTogram (CENTRIST). Experimental results revealed that CENTROG offers better detection and classification accuracy for both pedestrian and detection and vehicle type recognition.

The distortion observed in fisheye cameras has proven to be a persistent challenge for numerous state-of-the-art object detection algorithms, instigating the development of various techniques aimed at mitigating this issue. This study aims to evaluate various methods for mitigating distortion in fisheye camera footage and their impact on video object detection accuracy and speed. Using Python, OpenCV, and third-party libraries, the researchers modified and optimized said methods for video input and created a framework for running and testing different distortion correction methods and object detection algorithm configurations. Through experimentation with different datasets, the study found that undistorting the image using the longitude-latitude correction with the YOLOv3 object detector provided the best results in terms of accuracy (PASCAL: 68.9%, VOC-360: 75.1%, WEPDTOF: 15.9%) and speed (38 FPS across all test sets) for fisheye footage. After measuring the results to determine the best configuration for video object detection, the researchers also developed a desktop application that incorporates these methods and provides real-time object detection and tracking functions. The study provides a foundation for improving the accuracy and speed of fisheye camera setups, and its findings can be valuable for researchers and practitioners working in this field.

. The proposals of this network are further evaluated using a Boosted Decision Trees classifier in order to reduce potential false positive detections. With a log-average miss rate of 29.83 % on the reasonable test set of the KAIST Multispectral Pedestrian Detection Benchmark, we improve the current state-of-the-art by about 18 %.

The support vector machine (SVM), an assuring new method for the classification, has been widely used in many areas efficiently. However, the online learning issue of SVM is still not addressed satisfactorily since when a new sample arrives to retrain the SVM to adjust the classifier. This may not be feasible for real-time applications due to the expensive computation cost for retraining the SVM. This paper propose an Online SVM classifier algorithm known as OSVM-CH, which is based on the convex hull vertices selection depends on geometrical features of SVM. From the theoretical point of view, the first d+1(d is the dimension of the input samples) selected samples are proved to be vertices of the convex hull. This guarantees that the selected samples in our method keep the greatest amount of information of the convex hull. From the pedestrian detection application point of view, the new algorithm can update the classifier without reducing its classification performance.

Nowadays, many pedestrians are injured or killed in traffic accidents. As a result, several artificial vision solutions based on pedestrian detection have been developed to assist drivers and reduce the number of accidents. Most pedestrian detection techniques work well on sunny days and provide accurate traffic data. However, detection decreases dramatically in rainy conditions. In this paper, a new pedestrian detection system (PDS) based on generative adversarial network (GAN) module and the real-time object detector you only look once (YOLO) v3 is proposed to mitigate adversarial weather attacks. Experimental evaluations performed on the VOC2014 dataset show that our proposed system performs better than models based on existing noise reduction methods in terms of accuracy for weather situations.

One of the most widely used advanced driver assistance systems (ADAS) for preventing pedestrian-vehicle collisions is the intersection collision warning system (ICWS). Most previous ICWSs have been implemented with in-vehicle distance sensors, such as radar and lidar. However, the existing ICWSs show some weaknesses in alerting drivers at intersections because of limited detection range and field-of-view. Furthermore, these ICWSs have difficulties in identifying the pedestrian's crossing intention because the distance sensors cannot capture pedestrian characteristics such as age, gender, and head orientation. To alleviate these defects, this study proposes a novel framework for vision sensor-based ICWS under a cloud-based communication environment, which is called the intersection pedestrian collision warning system (IPCWS). The IPCWS gives a collision warning to drivers approaching an intersection by predicting the pedestrian's crossing intention based on various machine learning models. With real traffic data extracted by image processing in the IPCWS, a comparison study is conducted to evaluate the performance of the IPCWS in relation to warning timing. The comparison study demonstrates that the IPCWS shows better performance than conventional ICWSs. This result suggests that the proposed system has a great potential for preventing pedestrian-vehicle collisions by capturing the pedestrian's crossing intention.

One of the most widely used advanced driver assistance systems (ADAS) for preventing pedestrian-vehicle collisions is the intersection collision warning system (ICWS). Most previous ICWSs have been implemented with in-vehicle distance sensors, such as radar and lidar. However, the existing ICWSs show some weaknesses in alerting drivers at intersections because of limited detection range and field-of-view. Furthermore, these ICWSs have difficulties in identifying the pedestrian's crossing intention because the distance sensors cannot capture pedestrian characteristics such as age, gender, and head orientation. To alleviate these defects, this study proposes a novel framework for vision sensor-based ICWS under a cloud-based communication environment, which is called the intersection pedestrian collision warning system (IPCWS). The IPCWS gives a collision warning to drivers approaching an intersection by predicting the pedestrian's crossing intention based on various machine learning models. With real traffic data extracted by image processing in the IPCWS, a comparison study is conducted to evaluate the performance of the IPCWS in relation to warning timing. The comparison study demonstrates that the IPCWS shows better performance than conventional ICWSs. This result suggests that the proposed system has a great potential for preventing pedestrian-vehicle collisions by capturing the pedestrian's crossing intention.

The first step in order to achieve low-level multi-sensor fusion is the registration of images from multiple types of sensors. This is a very important task: it can be useful to improve the detection or the tracking of a moving object in an area. Putting together the information of an IR (infrared) and a visual camera we can use the information of the heat emanated from a human body to detect a pedestrian in the video. Basically we can align the IR and visual data knowing the calibration of the sensors, and always moving them together. In a real situation, it can be useful to align the images without imposing anything on the starting condition of the cameras and their relative position. In this paper, we present a method to automatically register IR with visual image data. The method uses geometric structures that are matched with a partial graph matching algorithm. We also introduce an iterative method to map IR and visual sequences using the homography matrix between frames. This method can be used to improve the multi-sensor motion detection: from an initial detection of a moving object in the visual image we can obtain the corresponding region in the thermal image.

Abstract—Computer vision techniques applied to systems used on road maintenance, which are related either to traffic signs or to the road itself, are playing a major role in many countries because of the higher investment on public works of this kind. These systems are able to collect a wide range of information au-tomatically and quickly, with the aim of improving road safety. In this context, the correct visibility of traffic signs and panels is vital for the safety of drivers. This paper describes an approach to the VISUAL Inspection of Signs and panEls (“VISUALISE”), which is an automatic inspection system, mounted onboard a vehicle, which performs inspection tasks at conventional driving speeds. VISUALISE allows for an improvement in the awareness of the road signaling state, supporting planning and decision making on the administration’s and infrastructure operators ’ side. A description of the main computer vision techniques and some experimental results obtained from thousan...

This paper introduces a novel benchmark to study the impact and relationship of built environment elements on pedestrian collision prediction, intending to enhance environmental awareness in autonomous driving systems to prevent pedestrian injuries actively. We introduce a built environment detection task in large-scale panoramic images and a detection-based pedestrian collision frequency prediction task. We propose a baseline method that incorporates a collision prediction module into a state-ofthe-art detection model to tackle both tasks simultaneously. Our experiments demonstrate a significant correlation between object detection of built environment elements and pedestrian collision frequency prediction. Our results are a stepping stone towards understanding the interdependencies between built environment conditions and pedestrian safety.

Human Motion detection is a challenging task due to a number of factors including variable appearance, posture and a wide range of illumination conditions and background. So, the first need of such a model is a reliable feature set that can discriminate between a human and a non-human form with a fair amount of confidence even under difficult conditions. By having richer representations, the classification task becomes easier and improved results can be achieved. The Aim of this paper is to investigate the reliable and accurate human motion detection models that are able to detect the human motions accurately under varying illumination levels and backgrounds. Different set of features are tried and tested including Histogram of Oriented Gradients (HOG), Deformable Parts Model (DPM), Local Decorrelated Channel Feature (LDCF) and Aggregate Channel Feature (ACF). However, we propose an efficient and reliable human motion detection approach by combining Histogram of oriented gradients (...

Pedestrian tracking video-based systems present particular problems such as the multi fragmentation or low level of compactness of the resultant blobs due to the human shape or movements. This paper shows how to improve the segmentation stage of a video surveillance system by adding morphological post-processing operations so that the subsequent blocks increase their performance. The adjustment of the parameters that regulate the new morphological processes is tuned by means of Evolution Strategies. Finally, the paper proposes a group of metrics to assess the global performance of the surveillance system. After the evaluation over a high number of video sequences, the results show that the shape of the tracks match up more accurately with the parts of interests. Thus, the improvement of segmentation stage facilitates the subsequent stages so that global performance of the surveillance system increases.

We introduce the new audio analysis task of pedestrian detection and present a new large-scale dataset for this task. While the preliminary results prove the viability of using audio approaches for pedestrian detection, they also show that this challenging task cannot be easily solved with standard approaches.

We introduce the new audio analysis task of pedestrian detection and present a new large-scale dataset for this task. While the preliminary results prove the viability of using audio approaches for pedestrian detection, they also show that this challenging task cannot be easily solved with standard approaches.

Object detection and tracking with the aid of computer vision is a most challenging task in the context of Driver Assistant System (DAS) for vehicles. This paper presents pedestrians detection techique using Haar-Like Features. The main aim of this research is to develop a detection system for vehicle drivers that will intimate them in advance for pedestrian's movement when they are crossing the zebra region or passing nearby to it along the road. For this purpose, dataset of 1000 images have been taken via CCTV camera which was mounted for road monitoring. A Haar based cascade classifiers have been implemented over images. And system is trained for positive (with people) and negative (without people) image samples, respectively. After testing, the obtained results show that it attained 90% accuracy while pedestrian detection. The proposed work provides significant contribution in order to reduce the road accidents as well as ensure the safety measurement for road management.

In the realm of Advanced Driving Assistance Systems (ADAS), the accurate assessment of pedestrian proximity is of paramount importance. This paper introduces a qualitative methodology that integrates YOLOv7-pose for object detection and pose estimation with the MiDaS (Monocular Depth Estimation in Real-Time with Deep Learning) model for monocular depth estimation. The main objective is to qualitatively assess pedestrian proximity to the camera within the ADAS framework. This procedure involves classifying pedestrians as "near" or "far" based on an inverse depth threshold that has been predetermined. In addition, the paper performs a qualitative comparative analysis of the results produced by the MiDaS Small, Hybrid, and Large variants to learn more about the performance of depth estimation in these contexts, particularly in relation to the presence of pedestrians. The evaluation emphasises this approach's qualitative potential for achieving situationally appropriate and context-aware pedestrian proximity assessment. The safety and adaptability of ADAS systems can be improved with the help of such insights, which have numerous applications in robotics, surveillance, and autonomous vehicles.

Late fusion schemes with deep learning classification patterns set up with multi-modality images have an essential role in pedestrian protection systems since they have achieved prominent results in the pedestrian recognition task. In this paper, the late fusion scheme merged with Convolutional Neural Networks (CNN) is investigated for pedestrian recognition based on the Daimler stereo vision data sets. An independent CNN-based classifier for each imaging modality (Intensity, Depth, and Optical Flow) is handled before the fusion of its probabilistic output scores with a Multi-Layer Perceptron which provides the recognition decision. In this paper, we set out to prove that the incremental cross-modality deep learning approach enhances pedestrian recognition performances. It also outperforms state-of-the-art pedestrian classifiers on the Daimler stereovision data sets.

Predicting pedestrian behavior is the key to ensure safety and reliability of autonomous vehicles. While deep learning methods have been promising by learning from annotated video frame sequences, they often fail to fully grasp the dynamic interactions between pedestrians and traffic, crucial for accurate predictions. These models also lack nuanced common sense reasoning. Moreover, the manual annotation of datasets for these models is expensive and challenging to adapt to new situations. The advent of Vision Language Models (VLMs) introduces promising alternatives to these issues, thanks to their advanced visual and causal reasoning skills. To our knowledge, this research is the first to conduct both quantitative and qualitative evaluations of VLMs in the context of pedestrian behavior prediction for autonomous driving. We evaluate GPT-4V(ision) on publicly available pedestrian datasets: JAAD and WiDEVIEW. Our quantitative analysis focuses on GPT-4V's ability to predict pedestrian behavior in current and future frames. The model achieves a 57% accuracy in a zero-shot manner, which, while impressive, is still behind the state-of-the-art domain-specific models (70%) in predicting pedestrian crossing actions. Qualitatively, GPT-4V shows an impressive ability to process and interpret complex traffic scenarios, differentiate between various pedestrian behaviors, and detect and analyze groups. However, it faces challenges, such as difficulty in detecting smaller pedestrians and assessing the relative motion between pedestrians and the ego vehicle.

An automated approach to explore the fundamental properties of high-density pedestrian traffic is outlined. The framework operates on video or time lapse images captured from surveillance cameras. For pedestrian velocity extraction, the framework incorporates cross-correlation based Particle Image Velocimetry (PIV) techniques. For pedestrian density estimation, the framework relies on the Machine Learning technique of the Boosted Regression Trees. The information collected from images in pixel coordinates are transformed to world coordinates with a pin-hole camera based projective transformation technique. The framework has been tested with high density crowd images acquired during the Muslim religious event, the Hajj. Accuracy and performance of the framework are reported.

Modern cars are being equipped with powerful computational resources for autonomous driving systems (ADS) as one of their major parts to provide safer travels on roads. High accuracy and real-time requirements of ADS are addressed by HW/SW co-design methodology which helps in offloading the computationally intensive tasks to the hardware part. However, the limited hardware resources could be a limiting factor in complicated systems. This paper presents a dynamically reconfigurable system for ADS which is capable of real-time vehicle and pedestrian detection. Our approach employs different methods of vehicle detection in different lighting conditions to achieve better results. A novel deep learning method is presented for detection of vehicles in the dark condition where the road light is very limited or unavailable. We present a partial reconfiguration (PR) controller which accelerates the reconfiguration process on Zynq SoC for seamless detection in real-time applications. By partially reconfiguring the vehicle detection block on Zynq SoC, resource requirements is maintained low enough to allow for the existence of other functionalities of ADS on hardware which could complete their tasks without any interruption. Our presented system is capable of detecting pedestrian and vehicles in different lighting conditions at the rate of 50fps (frames per second) for HDTV (1080x1920) frame.

in recent years, the number of people killed on roads has increased enormously, several pedestrian detection techniques in monocular images have been proposed to address this problem. We present our pedestrian protection system from moving vehicles using video cameras installed on the vehicle, this system combines pedestrian detection, trajectory estimation, risk evaluation, and driver alert. First, we focus on the pedestrian recognition task. Different combinations of image descriptors and classification methods have been evaluated on this task. Experiments are performed on a dataset captured on-board a vehicle driving through urban environments. Results show that the best model is HOG&RbfSVM.

Pedestrian detect plays a crucial role in security, intelligent surveillance, vehicles, and robotics. Occlusion handling is a challenging worry in tracking multiple people. The tracking is based on the highest accuracy object detectors. In the current paper, we proposed a framework that detects multiple pedestrians in the image, which depends on Faster Regionbased Convolutional Neural Network (R-CNN). We applied the transfer learning concept by using the VGG19 & VGG16 deep networks, which are trained before on Image-Net to extract the feature map. Relying on trained weights, to reduce the time of training, we used the transfer learning concept. The framework was tested on Penn-Fudan pedestrian database. The pedestrian detection accuracy was measured by using the area under the curve (AUC) of the receiver operating characteristic (ROC) that e is achieved 95.6%. In addition, the proposed system achieved Miss Rate (MR) equals 1.98, accuracy (ACC) equals 97.31%, and F1-score equals 93.17%. The achieved results show the promise of our proposed technique to detect multiple pedestrians in a single scene.

Along with the development of powerful processing platforms, heterogeneous architectures are nowadays permitting new design space explorations. In this paper we propose a novel heterogeneous architecture for reliable pedestrian detection applications. It deploys an efficient Histogram of Oriented Gradient pipeline tightly coupled with a neuro-inspired spatio-temporal filter. By relying on hardware-software co-design principles, our architecture is capable of processing video sequences from real-word dynamic environments in real-time. The paper presents the implemented algorithm and details the proposed architecture for executing it, exposing in particular the partitioning decisions made to meet the required performance. A prototype implementation is described and the results obtained are discussed with respect to other state of the art solutions.

Highways are an important component of any country. However, some highways in Indonesia endanger users while maintaining road safety. Crack detection early in the deterioration process can prevent further damage and lower maintenance costs. A recent study sought to develop a method for detecting road damage by combining the road damage detection (RDD) dataset with generative adversarial network technology and data augmentation to improve training. The current study aims to broaden the you only look once (YOLO) framework by incorporating the Swin Transformer into the chiral stationary phases (CSP) component of YOLOv7, with the goal of improving object detection accuracy in a variety of visual scenarios. The study compares the performance of various object detection models with varying parameters and configurations, such as YOLOv5l, YOLOv6l, YOLOv7-tiny, YOLOv7, and YOLOv7x. YOLOv5l has 46 million parameters and 108 billion floating point operations per second (FLOPS), whereas YOLOv6l has 59.5 million parameters and 150 billion FLOPS. With 31 million parameters and 140 billion FLOPS, the YOLOv7-swin model performs best with mean average precision (mAP), mAP_0.50 of 0.47. and mAP_0.5:0.95 of 0.232. The experimental results show that our YOLOv7-swin model outperforms both YOLOv7x and YOLOv7-tiny. The proposed model significantly improves object detection accuracy while keeping complexity and performance in balance.

Object detection, one of the most fundamental and challenging problems in computer vision. Nowadays some dedicated embedded systems have emerged as a powerful strategy for deliver high processing capabilities including the NVIDIA Jetson family. The aim of the present work is the recognition of objects in complex rural areas through an embedded system, as well as the verification of accuracy and processing time. For this purpose, a low power embedded Graphics Processing Unit (Jetson Nano) has been selected, which allows multiple neural networks to be run in simultaneous and a computer vision algorithm to be applied for image recognition. As well, the performance of these deep learning neural networks such as ssd-mobilenet v1 and v2, pednet, multiped and ssd-inception v2 has been tested. Moreover, it was found that the accuracy and processing time were in some cases improved when all the models suggested in the research were applied. The pednet network model provides a high performanc...

Important gains have recently been obtained in object detection by using training objectives that focus on hard negative examples, i.e., negative examples that are currently rated as positive or ambiguous by the detector. These examples can strongly influence parameters when the network is trained to correct them. Unfortunately, they are often sparse in the training data, and are expensive to obtain. In this work, we show how large numbers of hard negatives can be obtained automatically by analyzing the output of a trained detector on video sequences. In particular, detections that are isolated in time, i.e., that have no associated preceding or following detections, are likely to be hard negatives. We describe simple procedures for mining large numbers of such hard negatives (and also hard positives) from unlabeled video data. Our experiments show that retraining detectors on these automatically obtained examples often significantly improves performance. We present experiments on multiple architectures and multiple data sets, including face detection, pedestrian detection and other object categories.

Advances in information and signal processing technologies have a significant impact on autonomous driving (AD), improving driving safety while minimizing the efforts of human drivers with the help of advanced artificial intelligence (AI) techniques. Recently, deep learning (DL) approaches have solved several real-world problems of complex nature. However, their strengths in terms of control processes for AD have not been deeply investigated and highlighted yet. This survey highlights the power of DL architectures in terms of reliability and efficient real-time performance and overviews state-of-theart strategies for safe AD, with their major achievements and limitations. Furthermore, it covers major embodiments of DL along the AD pipeline including measurement, analysis, and execution, with a focus on road, lane, vehicle, pedestrian, drowsiness detection, collision avoidance, and traffic sign detection through sensing and vision-based DL methods. In addition, we discuss on the performance of several reviewed methods by using different evaluation metrics, with critics on their pros and cons. Finally, this survey highlights the current issues of safe DL-based AD with a prospect of recommendations for future research, rounding up a reference material for newcomers and researchers willing to join this vibrant area of Intelligent Transportation Systems.

Despite the rapid development of technologies for the detection of pedestrians in traffic, there are still significant problems with automated vehicle management systems. Often these problems are caused by different external conditions, weather disasters, different light levels, day and night, and more. All these factors can affect the incorrect assessment of the system and the performed detection of the pedestrian object. Pedestrian fatalities have risen recently, even as vehicles are equipped with sophisticated avoidance technology. Vehicle-based sensors can fail to identify pedestrians especially when those pedestrians are small, too far or too close to the vehicle, or partially occluded by nearby objects. Pedestrians have variable physical characteristics and appear in a variety of environments with different background features. The paper analyzes the possibilities and problems that arise in the detection and recognition of pedestrians. Pedestrian safety is one of the important...

The requirements demanded for pedestrian crossings, as the meeting point of pedestrian and vehicle traffic, are becoming more and more numerous and to guarantee that they are understandable, employable and feasible for all people in safe and convenient conditions, and in the most natural and autonomous manner (LEY 51, 2003). Natural stone is a highly demanded material used for covering and surface finishing of public external spaces. However, complying with the increasingly demanding requirements for pedestrian crossings, makes using natural stone much more difficult due to the inherent limitations that this material presents: from the labour-intensive mechanical procedures required to modify its shape, (as it cannot be moulded), to the inability to change its properties by chemical addition. This situation has been the premise of this investigation, with the objective of establishing a verification model of the functionalities of the pedestrian crossing, in general, and in particul...

Objects such as pedestrians exhibit large intra-class variations, posing significant challenges for visual object detection. State-of-the-art part-based models explicitly model object deformations, but are limited in their ability to handle image variations incurred by other geometric and photometric changes, such as human pose, lighting, occlusions, and large appearance variations. In this paper, we propose a novel approach which uses a spatially-biased hierarchical scheme to map features into a high-dimensional space that better represents the rich set of object appearance and local deformation variations. We propose a new algorithm to jointly learn the classification function and feature pooling in this high-dimensional space, in a structured prediction setting. Our approach achieves the best detection performance on the INRIA pedestrian dataset.

Counting the number of pedestrians in urban environments has become an area of interest over the past few years. Its applications include studies to control vehicular traffic lights, urban planning, market studies, and detection of abnormal behaviors. However, these tasks require the use of intelligent algorithms of high computational demand that need to be trained in the environment being studied. This article presents a novel method to estimate pedestrian flow in uncontrolled environments by using the fractal dimension measured through the box-counting algorithm, which does not require the use of image pre-processing and intelligent algorithms. Four scenarios were used to validate the method presented in this article, of which the last scene was a low-light surveillance video, showing experimental results with a mean relative error of 4.92% when counting pedestrians. After comparing the results with other techniques that depend on intelligent algorithms, we can confirm that this method achieves improved performance in the estimation of pedestrian traffic.

In pedestrian detection applications, one of the most popular frameworks that has received extensive attention in recent years is widely known as a 'Hough forest' (HF). To improve the accuracy of detection, this paper proposes a novel split function to exploit the statistical information of the training set stored in each node during the construction of the forest. The proposed split function makes the trees in the forest more robust to noise and illumination changes. Moreover, the errors of each stage in the training forest are minimized using a global loss function to support trees to track harder training samples. After having the forest trained, the standard HF detector follows up to search for and localize instances in the image. Experimental results showed that the detection performance of the proposed framework was improved significantly with respect to the standard HF and alternating decision forest (ADF) in some public datasets.