Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 14 views

    Vision-Based Vehicle Classification Using Background Subtraction and Deep Learning Model

    Uploaded by

    Ahad rahaman
    The document proposes developing a vision-based vehicle classification system using background subtraction and a deep learning model. Video data would be collected on highways in Bangladesh and vehicles would be detected and extracted. A dataset would be prepared and used to train and validate a convolutional neural network model for classifying vehicles. The model performance would be evaluated and compared to other machine learning methods. The system could provide more accurate traffic data and potentially be integrated into traffic management systems.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd

    Vision-Based Vehicle Classification Using Background Subtraction and Deep Learning Model

    Uploaded by

    Ahad rahaman
    14 views16 pages
    The document proposes developing a vision-based vehicle classification system using background subtraction and a deep learning model. Video data would be collected on highways in Bangladesh and vehicles would be detected and extracted. A dataset would be prepared and used to train and validate a convolutional neural network model for classifying vehicles. The model performance would be evaluated and compared to other machine learning methods. The system could provide more accurate traffic data and potentially be integrated into traffic management systems.

    Original Title

    Untitled

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document proposes developing a vision-based vehicle classification system using background subtraction and a deep learning model. Video data would be collected on highways in Bangladesh and vehicles would be detected and extracted. A dataset would be prepared and used to train and validate a convolutional neural network model for classifying vehicles. The model performance would be evaluated and compared to other machine learning methods. The system could provide more accurate traffic data and potentially be integrated into traffic management systems.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    14 views16 pages

    Vision-Based Vehicle Classification Using Background Subtraction and Deep Learning Model

    Uploaded by

    Ahad rahaman
    The document proposes developing a vision-based vehicle classification system using background subtraction and a deep learning model. Video data would be collected on highways in Bangladesh and vehicles would be detected and extracted. A dataset would be prepared and used to train and validate a convolutional neural network model for classifying vehicles. The model performance would be evaluated and compared to other machine learning methods. The system could provide more accurate traffic data and potentially be integrated into traffic management systems.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    of 16

    VISION-BASED VEHICLE CLASSIFICATION USING BACKGROUND

    SUBTRACTION AND DEEP LEARNING MODEL

    Presented By: Supervised By:

    Zeba Fariha Dr. Md. Hadiuzzaman


    Student ID: 0421042414 (F) Professor,
    Department of Civil Engineering,
    Bangladesh University of Engineering &
    Technology

    Bangladesh University of Engineering & Technology, Dhaka-1000, Bangladesh


    OUTLINE

     BACKGROUND & PRESENT STATE OF THE PROBLEM


     OBJECTIVES
     POSSIBLE OUTCOME
     METHODOLOGY
     COST ESTIMATION
    BACKGROUND & PRESENT STATE OF THE PROBLEM

     Intelligent Transportation System (ITS) is broadly noticeable for its innovations in different
    sectors of Transportation Engineering.

     Vehicle classification is one of the salient features of ITS.

     Deep learning (DL) is a sub-field of artificial neural networks which outperforms the
    conventional methods such as laser and loop induction technology and seems advantageous in
    terms of accuracy.
    BACKGROUND & PRESENT STATE OF THE
    PROBLEM

     Traffic management in Bangladesh lacks efficiency because of weak traffic control, monitoring
    and poor law enforcement. Traffic surveillance tasks i.e., identifying, classifying, monitoring
    and control of traffic are performed by manual observation.

     DL algorithm needs a well-range dataset. A new dataset from the perspective of Bangladesh as
    well as a DL based vehicle classification model has to be developed to replace manual vehicle
    classification approach.
    OBJECTIVES

    Main Objective:
    To develop a model that categorizes the vehicles according to their classes.

    Specific Objectives:
     Segmentation of vehicular images from captured videos using State-of-Art vehicular detection
    technique and aggregation of these images into a dataset.
     Developing Deep Learning (DL) based vehicle classification model to classify the vehicles from video.
     Calibrating and validating the DL model using the dataset and evaluating the performance of the model
    using different measure of performance.
     Comparing the calibrated DL model with other Machine Learning (ML) models used in different image
    classification algorithm.
    POSSIBLE OUTCOME

     More accurate and reliable traffic data and estimation of traffic state.
     Data can be transferred to a cloud storage and opens a new door for cloud computing.
     Vehicle identification and classification module of Vehicle Actuated Signal.
     Additional sensory module in toll collection management system.
    METHODOLOGY

    Vehicle Detection
    Dataset Deep Learning Vehicle
    Data Collection & Image
    Preparation Model Classification
    Extraction

    Figure 1– The layout of the proposed approach


    METHODOLOGY
    Data Collection

     A minimum of 12 hours video (8 am- 12pm, 2pm-6pm and 6 pm-10 pm) data of 3 days will be
    collected using HD night vision camera that will result into an approx. 60 million frames.
     The video will be captured from a raised platform (minimum 20 feet) on a firm ground to avoid
    occlusion and camera jitter.
     in 2 camera positions
     at 3 locations
     Data will be collected from Rayerbag, Matuail and Shanarpar located on Dhaka-Chittagong
    Highway (N1).
     Covers both vehicles from RHD and BRTA vehicle classification
     Optimum video capturing environment
    METHODOLOGY

    Vehicle Detection & Image Extraction


     Vehicle detection will be performed by feeding the collected video data into a state of art
    tool.

     The mechanism of this tool is vehicle detection and extraction in image format. Vehicle
    detection is followed by 2 steps:

    Background Foreground Feature


    subtraction Extraction
    METHODOLOGY

     Background Subtraction: Foreground objects(vehicle) will be segmented from the input


    images through this technique. The binary image will be estimated using threshold-
    difference function.

     Foreground Feature Extraction Connected component labeling will be used to collect


    and merge foreground pixels into objects by judging their eight-connectivity relationship
    with the adjacent neighboring pixels. Afterwards, the component properties (i.e., Bounding
    box) will be extracted from the geometry of the detected object.
    METHODOLOGY

    Dataset Preparation

     Images of the detected vehicles will be extracted from the frames incorporating the
    bounding box.
     After manual labeling and classifying the images, a dataset will be prepared.
    METHODOLOGY

    Deep Learning Model

     A suitable DL structure (i.e., Convolutional Neural Network (CNN)) will be adopted.


     Train the structure with the dataset by calibrating and validating its hyperparameters.
     The final output will give the classification of input vehicle images.
     Comparison will be done with other ML models such as Support Vector Machine (SVM)
    and neural networks i.e.,YOLOv4.
     As a coding platform MATLAB will be used along with Deep Learning Toolbox.
    COST ESTIMATION

    Sl. Sector Item Name Unit Unit Price


    No Price (Tk.)
    (Tk.)

    1 Data Collection Hiring 2 HD night vision 3 days, 2 7500 45000


    cameras and personnel for positions
    video data collection

    2 Typing, Drafting, Binding & Paper etc. - - 4500

    Total 49500
    REFERENCES
    [1] Wang, Xinchen , Zhang, Weiwei , Wu, Xuncheng , Xiao, Lingyun , Qian, Yubin , Fang, Zhi . "Real-time vehicle type classification with deep convolutional neural
    networks." Journal of Real-Time Image Processing, Springer, Vol 16, pp. 5-14, Berlin Heidelberg, 10.1007/s11554-017-0712-5. (2019).

    [2] Li, Yinghua , Song, Bin , Kang, Xu , Du, Xiaojiang , Guizani, Mohsen . "Vehicle-Type Detection Based on Compressed Sensing and Deep Learning in Vehicular
    Networks ." Sensors, mdpi, Vol 18, pp. 1-15, Switzerland, 10.3390/s18124500. (2018).

    [3] Hasan, Md Mahibul, Wang, Zhijie , Hussain, Muhammad Ather Iqbal, Fatima , Kaniz . "Bangladeshi Native Vehicle Classification Based on Transfer Learning with Deep
    Convolutional Neural Network." Sensors, mdpi, Vol 21, pp. 7545, Basel, Switzerland, https://doi.org/10.3390/s21227545. (2021).

    [4] Tabassum, Shaira , Ullah, Md. Sabbir, Al-nur, Nakib Hossain, Shatabda, Swakkhar . "Native Vehicles Classification on Bangladeshi Roads Using CNN with Transfer
    Learning." 2020 IEEE Region 10 Symposium (TENSYMP), IEEE, pp. 40-43 Dhaka, Bangladesh , 10.1109/TENSYMP50017.2020.9230991. 2020.

    [5] Onim, Md. Saif Hassan, Akash, Muhaiminul Islam, Haque, Mahmudul , Hafiz, Raiyan Ibne. "Traffic Surveillance using Vehicle License Plate Detection and Recognition
    in Bangladesh." 2020 11th International Conference on Electrical and Computer Engineering (ICECE), IEEE, pp. 121-124 Dhaka, Bangladesh,
    10.1109/ICECE51571.2020.9393109. 2020.

    [6] Maungmai, Watcharin , Nuthong, Chaiwat . "Vehicle Classification with Deep Learning." 2019 IEEE 4th International Conference on Computer and Communication
    Systems (ICCCS), IEEE, pp. 294-298 Singapore, https://doi.org/10.1109/ICCCS46626.2019. 2019.

    [7] Yu, Shaoyong , Wu, Yun , Li, Wei , Song, Zhijun , Zeng, Wenhua . "A model for fine-grained vehicle classification based on deep learning." Neurocomputing , Elsevier,
    Vol 257, pp. 97-103, 10.1016/j.neucom.2016.09.116. (2017).

    [8] Song, Huansheng , Liang, Haoxiang , Li, Huaiyu , Dai, Zhe , Yun, Xu . "Vision-based vehicle detection and counting system using deep learning in highway scenes."
    European Transport Research Review, Springer, Vol 11, pp. 1-16, https://doi.org/10.1186/s12544-019-0390-4. (2019).

    [9] Hadiuzzaman, Md. , Haque, Nazmul , Rahman, Fahmida , Hossain, Sanjana , Siam, Mohammad Rayeedul Kalam, Qiu, Tony Z.. "Pixel-based heterogeneous traffic
    measurement considering shadow and illumination variation." Signal, Image and Video Processing, Springer, Vol 11, pp. 1245–1252, London, 10.1007/s11760-017-1081-z.
    (2017).

    [10] Fernando, W. S. K., Herath, H. M. S. P. B., Perera, P. H., Ekanayake, M. P. B., Godaliyadda, G. M. R. I., Wijayakulasooriya, J. V.. "Object identification, enhancement
    and tracking under dynamic background conditions." 7th International Conference on Information and Automation for Sustainability, IEEE, pp. 1-6 Colombo, Sri Lanka,
    10.1109/ICIAFS.2014.7069583. 2014.

    You might also like