Research on Real-time Dynamic Object Detection Based on YOLOv3 Deep Learning Network
Yang Yu · 2023
Based on YOLOv3 deep learning network, this paper aims to realize real-time dynamic target detection. Dynamic target detection is essential in applications, especially video surveillance and autonomous driving. However, target detection requires processing large amounts of image data, and conventional algorithms need help to meet real-time requirements. Therefore, deep learning technology has become practical for achieving real-time dynamic target detection. This article is divided into the following sections. First, it introduces the structure and principle of the YOLOv3 network, including multi-scale feature extraction, anchor frame design, NMS, and other technologies. Second, we propose optimization methods to meet the demand for real-time dynamic target detection, such as convolution-based cross-frame target tracking and deep learning-based image enhancement. Finally, experiments using standard datasets show the proposed method's performance and robustness in real-time dynamic target detection. To sum up, this article aims to explore solutions in real-time dynamic target detection and provide references for practice.