Location via proxy:   
[Report a bug]   [Manage cookies]                
Salih Marangoz | Portfolio Website

Salih Marangoz

Computer Science M.Sc. Graduate


Enter Portfolio
Salih Marangoz

Salih Marangoz

Computer Science M.Sc. Graduate


I am Salih Marangoz, Computer Science M.Sc. graduate from the University of Bonn, focusing on Robotics (specifically perception and motion) and Machine Learning. In my past experiences, I worked on improving perception in agricultural robots, also navigation and mapping in search-and-rescue robots. I am also skilled in machine learning and deep learning with a strong C/C++ and Python background. I aim to build my career in developing intelligent systems.


Projects

×

Dawn IK: Decentralized collision-aware inverse kinematics solver for heterogeneous multi-arm systems

This project introduces a novel approach to compute inverse kinematics for serial manipulators that prioritizes smooth, collision-free motion in multi-robot environments. By formulating constraints as weighted cost functions optimized via non-linear solvers, the method avoids computationally expensive inverse Jacobian calculations and does not require redundant degrees of freedom. The system demonstrates superior collision avoidance and trajectory tracking compared to state-of-the-art methods, enabling decentralized control of multiple robotic arms in confined, overlapping workspaces without collisions.


Video
Source Code
Paper

Dawn IK: Decentralized collision-aware inverse ...

DawnIK is a real-time inverse kinematics solver for robotic arms focusing on observation capabilities with collision avoidance and multiple objectives.

×

Fruit Mapping with Shape Completion for Autonomous Crop Monitoring

This project tackles the challenges of autonomous crop monitoring by developing a system to map and estimate the shape of fruits, even in complex plant structures where occlusions occur. Using advanced image segmentation, 3D mapping, and superellipsoid fitting, the approach generates accurate fruit shape and volume estimates from partial data. Tested with a robotic arm and RGB-D camera, the system demonstrates high accuracy in simulated and real-world glasshouse environments, enabling future applications like yield estimation and automated fruit picking.


Video
Source Code
Paper

Fruit Mapping with Shape Completion for Autonom...

Superellipsoid shape registration on Capsicum fruits for predicting their position, orientation and sizes.

×

Depth Image Averager on GPU

This project focuses on depth image averaging as a simple yet effective method for improving depth measurements in static scenes. Unlike complex approaches like TSDFs, this technique requires the sensor to remain stationary to produce accurate results. By averaging depth images, the package addresses the efficient processing of depth data and reduces the high computational cost of registering every single depth image in the mapping system, enhancing performance in high-resolution settings or on slower hardware.



Source Code

Depth Image Averager on GPU

Pixel-wise depth image averaging and outlier filtering with OpenCL support.

×

Depth Image Veiling Effect (Shadow Noise) Filter

Veiling effect (shadow noise or ghost points) appears at the edge discontinuties in depth images where it becomes difficult to determine the depth value. This project provides a more efficient and faster way to filter these points compared to pcl::ShadowPoints and at least 20x faster than pcl::StatisticalOutlierRemoval.



Source Code

Depth Image Veiling Effect (Shadow Noise) Filter

A more efficient per-frame filtering approach to remove the ghost points appearing on edge discontinuties in depth images.

×

A* Pathfinder C++ Library

Path finder on 2D grids with A* algorithm with 400 lines of code in a single header file. Includes a stable (correct) and a fast (approximate) path finders for different needs.



Source Code

A* Pathfinder C++ Library

A single header library for path finding on 2D grids with A* algorithm. Includes a stable and a fast path finders.

×

Fast LaserScan Line Segment Extraction

This project solves the line segment extraction for low-powered robots. If you have more room for processing power then consider using salihmarangoz/laser_line_extraction (my fork of a good project where I solved the timestamp issue)


Video
Source Code

Fast LaserScan Line Segment Extraction

Fast line segment extraction using the Douglas-Peucker algorithm for low-powered robots.

Recent Blog Posts

Image Imputation using SVD

Previously, we used SVD (Singular Value Decomposition) for lossy image compression. But could the same approach be adapted to recover an image with missing pixels?
August 18, 2024

2D Scan Matcher using PyTorch

Exploring a 2D scan matcher implementation using PyTorch, focusing on solving the problem without explicitly computing correspondences for efficient and accurate results.
March 19, 2023

Efficient Personal Backup Strategies using BTRFS

My differential backup strategy using rsync and an external storage formatted with BTRFS+LUKS which supports encryption, on-the-fly compression, de-duplication, and filesystem-level snapshot.
July 13, 2022