基于深度学习的边境行为识别系统探讨

Exploration of Border Behavior Recognition System Based on Deep Learning

在视频监控领域,通过图像的变化能够实现移动侦测区域报警、人脸识别等,但功能相对单一,易出现误报和漏报等问题,需要人工参与分析识别,而且在边境视频监控系统中,视频采集路数较多,都用人工分析识别显然不现实。针对这种情况,提出了一种基于深度学习的多人边境行为识别方法。该方法通过建立特定的三种典型边境行为(走私、翻越拦阻和破坏拦阻)数据库,再利用计算结构简单的Light weight OpenPose及KNN分类算法,实现运用一种处理实时性好、分类准确率高的边境特定行为识别方法。该方法不仅算法结构简单,对计算资源要求低,而且具备对多人行为进行有效识别的能力,非常适合应用到边境视频监控系统中,提升边境的智能化水平。

In the field of video surveillance,changes in images can achieve mobile detection of area alarms,facial recognition,etc.However,the function is relatively single,prone to issues such as false alarms and missed alarms,and requires manual participation in analysis and recognition.Moreover,in border video surveillance systems,there are many video acquisition channels,and it is obviously not realistic to use manual analysis and recognition.A multi person border behavior recognition method based on deep learning is proposed to address this situation.This method establishes a database of three typical border behaviors(smuggling,crossing obstruction,and sabotage obstruction),and then utilizes the Light weight OpenPose and KNN classification algorithms with simple computational structures to achieve a border specific behavior recognition method with good real-time processing and high classification accuracy.This method not only has a simple algorithm structure and low computational resource requirements,but also has the ability to effectively identify multi-person behavior,It is very suitable for application in border video surveillance systems to enhance the intelligence level of the border.