自动扶梯乘客危险行为检测方法

Research on the detection method of escalator passengers’dangerous behaviors

文中主要针对当前方法研究者对自动扶梯乘客危险行为检测效率不足的问题,在行人逆行检测方法中提出改进的单目标检测算法,结合改进卡尔曼滤波及表观信息提取模块的DeepSort目标追踪算法构建乘客逆行检测模型提高行人逆行检测的准确率。在行人摔倒检测方法中利用轻量化改进OpenPose模块提取的乘客骨骼关键点,通过危险行为自动采集模块转化得到输入特征参数,采用小波去噪方法(WD)结合随机森林方法(RF)构建摔倒行为检测模型。结果表明,相较于未改进的单目标检测方法结合未改进的DeepSort目标追踪算法其行人逆行检测的准确率提高了3.5%。相较于SVM、KNN对摔倒行为进行检测的方法,轻量化改进的OpenPose模块与WD-RF相结合的行人摔倒检测准确率分别提升了2.5%、1.8%。

Considering that the current method is inefficient in detecting dangerous behaviors of escalator passengers,ISSD target detection algorithm is proposed in pedestrian retrograde detection method,and a passenger retrograde detection model was constructed by combining DeepSort target tracking algorithm with improved Kalman filtering and apparent information extraction module to improve the accuracy of pedestrian retrograde detection.In the pedestrian fall detection method,the key points of passengers’bones extracted by the lightweight improved OpenPose module were transformed into input characteristic parameters by the dangerous behavior automatic acquisition module,and the fall behavior detection model was constructed by wavelet denoising method(WD)combined with random forest method(RF).Results show that the accuracy of pedestrian retrograde detection is improved by 3.5%compared with the unmodified SSD method combined with the unmodified DeepSort target tracking algorithm.Compared with SVM and KNN,the accuracy of pedestrian fall detection by combining lightweight and improved OpenPose module with WD-RF is improved by 2.5%and 1.8%respectively.