This repository contains code samples
Is an implementation of Histogram of Oriented Gradient with Support Vector Machine in CUDA. Created for CUDA programming course at AGH UST.
Is a simple library to generate COCO format datasets out of known labels within CARLA simulation engine. Created for computer vision in autonomous vehicles course at AGH UST.
Raspberry pi pico MicroROS firmware with data collection and processing scripts for capstone project at AGH UST. Based on measured data and parameters from datasheet system is accurately identified. Later Kalman Filter, LQR, LQI and LQG controllers are examined on this system.
Short computer vision code samples created for computer vision in autonomous vehicles course at AGH UST. Meant to be run with CARLA simulation engine.
- ClassicDetector.py is a color-space-based approach of traffic light detection. Uses HSV and YCbCr color thresholding and basic geometry properties of expected object to be detected.
- faster_rcnn.ipynb is transfer learning of Faster RCNN in order to learn it to detect traffic lights. Trained using own dataset created with dataset_creator. Unfortunately performance was poor.
- svm_train.ipynb creates augumented dataset and trains Support Vector Machine on Histogram of Oriented Gradients.
- SVMDetector.py expands ClassicDetector.py by adding detection validation with HOG+SVM for better robustness.
collect.py and drive.py were created for dataset creating for HOG+SVM and for runtime evaluation of algorithms.