Research and Implementation of Embedded Real-time Target Detection Algorithm Based on Deep Learning

Abstract

In view of the poor detection effect and low robustness of traditional target detection algorithms, this paper studies target detection algorithms based on deep learning, and designs an embedded real-time target detection evaluation board based on AI chip. It is realized that most of the common deep convolutional neural network models represented by YOLOv3 can be transplanted to the board. Although there is a certain accuracy loss within the acceptable range, this implementation achieves a faster speed to meet the needs of real-time detection. Meanwhile, a complete set of video component calling and network transmission schemes are proposed. By designing a unified standard interface accessed to the program framework, the implementation board can be flexibly extended to meet the needs of various artificial intelligence applications.