Vehicle Automation Using YOLOv7 and Computer Vision
Juan George Thomas, Jimmy Dungdung, Keyur Kumar Kanwat, Avinash Ratre · 2023
In this paper we strive to achieve real-time vehicle and lane detection on low powered consumer hardware. The main goal of this work is to reduce the response time without sacrificing accuracy. For a system to be considered realtime the FPS should be above 25. This work addresses the low performance of previous systems by designing a pipeline and compiling libraries for it specific to the target machine. Previous works on this like Hybridnets produce good results but take up too much computing resources. For this paper we have used YOLOv7 as the core for vehicle detection. YOLOv7 is a state-of-the-art object detection framework released in July 2022. It will be custom trained on a traffic specific dataset for maximum accuracy. In addition to that a computer vision based lane intelligence system would be there to support it. The lane system will complement Yolo by predicting the lanes and calculating the threat of a collision from nearby vehicles. The pipeline serially runs the model and then the lane detection system. For the lane detection system OpenCV library was compiled on the target machine to maximize the efficiency.