Performance Evaluation of SOTA Object Detector YOLOv5 for Vehicle Detection on Edge Devices

One of the primary applications of computer vision is object detection. It requires large computation capabilities and on edge devices detection is a difficult operation. Due to low computation capabilities of edge devices, it is difficult to emulate the same performance as that of workstation. We have built a model for vehicle classification and counting, using State of Art object detector YOLOv5 to detect vehicles such as car, bus, truck, LCV, multi axle, auto and tractor, inference the object detector YOLOv5 model in terms of accuracy and time on different edge devices such as Raspberry pi4, Jetson Nano and hand-held Android Device. In this study we found that Jetson Nano is performing better than other edge devices for inferencing time on a single image.