Abstract
When it comes to edge applications that require object detection based on computer vision, there are several models that can be used. However there is not an easy way to identify what is the best model for each application. In this paper we present a com prehensive comparison of several YOLO-based models for object detection targeting edge applications using single board computers (Raspberry pi). We compare several versions of YOLO (YOLOv5, v8 and v10) in terms of accuracy, inference time and training time. For the specific comparison we evaluate the performance for an edge application where the system has to identify the empty parking spots in a parking lot. The comparison allows developers to select the best model based on the application requirements.
