Embedded Object Detection System Based on Deep Neural Network

Abstract

Object detection is widely used in many fields, such as intelligent security monitoring, smart city, power inspection, and so on. The object detection algorithm based on deep learning is a kind of storage intensive and computing intensive algorithm which is difficult to achieve on the embedded platform with limited storage and computing resources. In this paper, we choose mobinetv2, a lightweight neural network with few model parameters and strong feature extraction ability, to replace darknet53 as the backbone network of YOLOv3 algorithm. In addition, we use a model compression method based on channel pruning to compress the network model. This method compresses model to detecting objects on embedded ARM platform. Neon instruction and OpenMP technology are further used to optimize and accelerate the intensive computing of convolutional network, and finally achieve a real-time embedded object detection system.