Stride Safe: Real-Time Pedestrian Detection Across Landscapes with YOLOv5 on Jetson Nano

Pedestrian detection is a crucial task in advanced driver assistance systems and autonomous driving. In recent years, pedestrian detection systems have achieved higher levels of precision. However, the existing algorithms fail in real-time scenarios. Aiming at this problem, an enhanced pedestrian detection algorithm utilizing YOLOv5 is proposed, which can run faster without compromising precision and even in complex real-world scenes. The proposed model on the Caltech Pedestrian dataset achieved precision of 87.2%, mAP of 84.2%, and an inference speed is 17.5ms which is a reduction of 36.1 % compared to improved YOLOv4 and makes the model suitable for real-time application, however a reduction in performance is seen while detecting occluded pedestrians in crowded places. The model is implemented on hardware using NVIDIA Jetson Nano for the real-time pedestrian detection and results are good.