An Edge Computing Architecture for Object Detection

Edge computing services are contingent on several constraints. There is a requirement needed to provide a proper function, such as low latency, low energy consumption, and high performance. Object detection analysis involves high power resources, it is because of the need to process the images or videos. In this paper, the architecture of edge computing for object recognition is proposed, and the performance of the edge node is examined. The resources performance comparison on Raspberry Pi and Neural Compute Stick are inspected. This study combined the Neural Compute Stick (NCS) to enhance the ability of image processing on Raspberry Pi. Through the aid of NCS, the Raspberry Pi’s frames per second (FPS) is increased by four times when the object detection program is executed, and the energy consumption of the Raspberry Pi is also recorded.