基于改进SSD的实验室安全员异常行为检测仿真

Simulation of Anomalous Behavior Detection for Laboratory Safety Officer Based on Improved SSD

由于实验室安全员人数较多,使得监管难以全面覆盖,监督范围难以精准把握,从而导致异常行为检测误差较大。为了有效解决以上问题,提出一种基于改进单激发多框探测器算法(Single Shot MultiBox Detector, SSD)算法的异常行为检测方法。在传统SSD的基础上,添加附加特征提取层以及卷积预测层,计算待检测人员图像尺度,作为先验框架的初始输入值,设定缩放因子调整特征提取框大小,求解不同动作的尺度特征。引入融合交叉熵,按照安全员异常行为定义,建立损失函数,求解异常行为出现概率,实现对安全员异常行为有效检测。实验结果证明,所提方法的误检率低于0.5,且检测耗时在10s内,可有效实现对异常行为动作的检测。

Due to the large number of laboratory safety officers,it is difficult to comprehensively cover supervision and accurately grasp the scope of supervision,resulting in significant errors in detecting abnormal behaviors.In this article,a method of detecting abnormal behaviors based on an improved Single Shot MultiBox Detector(SSD)algorithm was put forward.Based on traditional SSD algorithm,an additional feature extraction layer and a convolutional prediction layer were added.Then,the scale of the image of the person to be detected was calculated as the initial input value of prior frame.Moreover,the scaling factor was established to adjust the size of feature extraction box,thus solving the scale features of different actions.Furthermore,a fused cross-entropy was introduced.Meanwhile,a loss function was established according to the definition of abnormal behaviors of safety officers.Finally,the probability of abnormal behavior was calculated.Thus,the effective detection of abnormal behavior of safety officer was achieved.The experimental results prove that the false detection rate of the proposed method is less than 0.5,and the detection only takes ten seconds.Therefore,the proposed method can effectively detect abnormal behaviors.