Object Detection in Autonomous Driving Systems Using YOLOv5 and Kalman Filtering

This paper describes an enhanced approach to perform object detection and tracking for the use of autonomous driving systems by using YOLOv5 model for the detection process, and Kalman filtering for tracking process. The objective is to overcome the difficult tasks of identifying and determining the position of several objects in the context of rapidly changing and diverse conditions that define urban roads, highways, parking lots and in adverse meteorological conditions. For object detection, the relatively accurate and fast YOLOv5 is applied, and for object tracking, we select Kalman filtering based on its stability in cases of occlusions or other changes between frames. The system is tested on several datasets including the urban, highway, parking lot, and adverse weather conditions and measures such as precision rate, recall rate, F1 rate and MOTA, and computational time are used to measure the performance of the algorithm. The findings show that the presented approach gains a high detection rate, consistent object tracking, and sufficiently fast processing, which can be effective for self-driven vehicles. Nevertheless, it has a slightly poor performance in low visibility complicated traffic condition, which is common in real-world application. [This work demonstrates how YOLOv5 and Kalman filtering can be used in autonomous driving and notes that such an approach provides a stable method for object detection and tracking in various conditions for autonomous driving]. In future related improvements, the performance of the developed system will be optimized for operation at even more severe conditions.