结合残差网络和TSM的暴力行为检测方法

A Violence Behavior Detection Method Combining Residual Networks and TSM

深度学习方法常用于辅助检测暴力行为,从而降低监控视频人工干预的依赖性。然而,随着深度网络的发展,梯度消失、过拟合等问题变得更加突出。为了解决这些问题,本文提出了一种结合残差网络和时间转移模块的方法,充分挖掘视频序列中的时空信息,优化动作识别效果,从而提高暴力行为检测的准确率。实验的结果显示,相较于直接使用残差网络ResNet50和ResNet101,本文方法对暴力行为的识别准确率分别提高了1.4%和0.7%。

Deep learning methods are commonly used to assist in detecting violent behavior,thereby reducing the dependence on manual intervention in surveillance videos.However,with the development of deep networks,problems such as vanishing gradients and overfitting have become more prominent.To address these issues,this paper proposes a method that combines residual networks and time transfer modules to fully explore the spatiotemporal information in frequency sequences,optimize action recognition performance,and improve the accuracy of violent behavior detection.The experimental results show that compared to directly using residual networks ResNet50 and ResNet101,our method improves the recognition accuracy of violent behavior by 1.4%and 0.7%,respectively.