DEEP LEARNING - BASED OBJECT DETECTION ON RASPBERRY PI

Abstract

With the recent development of deep learning, object detection methods have achieved high performance on both speed and accuracy on modern desktop systems. In addition, the development of smaller and faster deep learning models to fit IoT devices has been attracting much attention. This article explores the relevance of object detection models on the Raspberry Pi, a popular embedded computer that can be integrated into IoT systems to make work easy. We consider the effects of two modern object detection models, Single Shot Detector (SSD) and You Only Look Once (YOLO). The models will be evaluated based on the frame rate and the average precision in the phase of inference. The experimental results show the feasibility of the models on low-profile computers.