Traffic Lights Recognition Using Computer Vision on the Microcomputer Platform

This paper aims to present traffic light recognition during the daytime. The system operates in real-time, using computer vision methods with the OpenCV library assistance. Implemented is performed on Raspberry Pi microcomputer. From a computer vision point of view, the stated problem is related to the colour-based object detection. The system performs colour segmentation to find traffic light candidates. An object detection algorithm then examines candidates. We used cascade classifier and support vector machine, which are object detection methods based on machine learning. These methods are available in the OpenCV library to find the most appropriate solution allowing real-time execution on Raspberry Pi. The system was tested on the freely available dataset LISA as well as our own dataset. Apart from Raspberry Pi, we also used another more powerful microcomputer Rock Pi. Therefore, we were able to test our system on three different performance classes, which should give us better insight into the ability of current microcomputers to execute object detection algorithms in real-time traffic environment.