Autonomous Vehicle Pedestrian Detection & Traffic Sign Recognition Using YOLOv8

Autonomous Vehicles (AVs) present a viable way to overcome several traffic-related problems, such as congestion, pollution, and accidents. In order to safely navigate in urban environments, AVs require accurate perception systems capable of detecting dynamic objects, predicting their behaviour, and interpreting information from static objects. Therefore, this paper focuses on presenting a comparative analysis of both real-time pedestrian detection and accurate identification of traffic signs in an effort to develop a robust AV system that prioritises these two elements during real-time object detection and avoidance. This development will enhance both urban safety and efficiency, while ensuring reliable and accurate object detection through the use of the YOLO (You Only Look Once) algorithm. First, A set of traffic signs and object images as a dataset was collected from the German Traffic Sign Recognition Benchmark (GTSRB) and the Penn-Fudan Pedestrian dataset for pedestrians. Second, image processing techniques including conversion to greyscale, image segmentation, and normalisation were applied using OpenCV. Third, the image data were passed to the training phase of the YOLOv8 model and went through the training and hyperparameters tuning process. The results show a mean Average Precision (mAP) of 94% accuracy in traffic sign recognition and a pedestrian detection precision of 90.67%. The findings underscore the importance of continued exploration of advanced object detection methods for AVs, such as data augmentation, to improve both the adaptability and robustness of AV systems in dynamic environments.

Перевод пока недоступен