Traffic light detection and counter recognition using video images and artificial intelligence

Al-Kumaim, Hefdhallah Abdulatef Senan (2021) Traffic light detection and counter recognition using video images and artificial intelligence. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Traffic light detection and counter recognition using video images and artificial intelligence.pdf - Submitted Version Download (599kB)
	Text (Full text) Traffic light detection and counter recognition using video images and artificial intelligence.pdf - Submitted Version Restricted to Repository staff only Download (2MB)

Abstract

Traffic light control systems are widely used for the purpose of monitoring and controlling the automobiles flow through the junction of many roads. Smooth motion of cars in the public transportation routes is the aim of invention and installation of the traffic light. However, ADAS (Advanced Driver Assistance System) is a technology used in vehicles which helps drivers for driving. ADAS is required to be implemented in vehicles for the purpose of providing guidance for drivers and enhancing a better safety for vehicles and roads by minimizing human error which often occur at roads intersections. Besides, recognition of the countdown timer of traffic light provides an input to the system of autonomous vehicles technology of the remining time on the traffic light in order to have a more accurate system in term of controlling the vehicle’s speed. Therefore, this project aims to design a vision system for detecting and recognizing traffic light with the counter digits. The proposed method suppressed the background by annotating all the candidates from the dataset which contains 2600 as well as classifying them to their respective classes which was accomplished by using Cash Value Accumulation Test (CVAT). The annotation files were merged with the original images and preprocessed using Roboflow. after uploading the dataset Roboflow generates a link to be used in Google Colab to import the dataset for training purpose to be validated and tested further. The algorithm used for object detection and recognition is YOLOv5 algorithm The evaluation of the method used was tested by 100 frames per class. The experiment resulted in an excellent detection and recognition rates of which the system was trained to with overall confidence rate of object detection varies between 80% to 90% as well as the accuracy of testing dataset which achieved an average of 95% to 100%.

Actions (login required)

View Item