基于轻量孪生网络的目标跟踪算法

2021-12-19 07:38:40王勇张志腾王瑛
电脑知识与技术 2021年32期

王勇 张志腾 王瑛

摘要:行人目标跟踪是智能监控领域的一个重要课题。传统的目标跟踪技术,在跟踪精度上没有深度网络高,但深度卷积神经网络计算量极大,导致计算速度缓慢无法实时跟踪。随着卷积网络的不断发展,孪生网络在目标跟踪这一课题上脱颖而出,其根据子网共享权重的特点,可以训练出有效的网络只需要少量的参数,少量的参数也就意味着不易于过拟合以及运行速度快等突出的优点,适用于实时行人目标跟踪。文中采用孪生网络和轻量骨干网络构建目标跟踪网络,实现实时高精度的目标跟踪算法。

关键词;目标跟踪;孪生网络;轻量网络;实时跟踪

中图分类号:TP18      文献标识码:A

文章编号:1009-3044(2021)32-0001-03

Object Tracking Algorithm Based on Lightweight Siamese Network

WANG Yong,  ZHANG Zhi-teng,  WANG Ying

(School of Computers, Guangdong University of Technology, Guangzhou 510006, China)

Abstract: Pedestrian object tracking is an important subject in the field of intelligent monitoring. The traditional object tracking technology is not as high as the deep network in tracking accuracy, but the deep convolutional neural network has a huge amount of computation, which leads to slow computing speed and unable to track in real time With the continuous development of convolution network, siamese network in object tracking on the subject, according to the characteristics of the subnet Shared weight, can train the effective network only need a small amount of parameters, a small amount of means is not easy to fitting parameters and running speed of such outstanding advantages, suitable for real-time object tracking of pedestrians. In this paper, the siamese network and the lightweight backbone network are used to construct the target tracking network, and the real-time and high precision target tracking algorithm is realized.

Key words:object tracking;siamese network;lightweight network;real-time tracking

现代社会智能化发展飞快,越来越多的监控摄像头以及视频图像分析技术应用于我们的日常生活中[1]。如商场中使用监控摄像头来关注分析顾客行为以防止顾客破坏超市的公共秩序以及道路上的红绿灯处监控摄像头用于监控来往车辆是否违反交通规则以此来约束司机不当行车行为,维护交通秩序。未来监控视频分析技术还可应用于大量聚集人群行为监控分析,实时分析人群行为,当发生暴动及斗殴等不良行为时自动报警以防止大规模冲突斗殴事件发生时未能及时发现并出警的情况发生。

由此可见,监控摄像头将会越来越普及,安装至各个公共场合甚至于各家各户中,其硬件成本随着时间的推移也会越来越低,视频分析技术也将会越来越精进,此时大商场中遍布大量摄像头将是一种必然的趋势。……

登录APP查看全文