Research On Recognition Method Of Gesture Actions Of Urban Rail Train Drivers

Nowadays,with the rapid development of urban rail transit,to ensure the safe operation of trains,not only lies in the timely maintenance of infrastructure diseases in the rail transit system,but also the professional and mobility of relevant core staff such as train drivers is particularly important.Whether a driver can make a correct gesture at such important nodes as the arrival,departure and closing of a train is an important standard to measure the working attitude and quality of a driver.Incorrect gesture judgment will directly threaten the safety of train operation,so it is very important to monitor and recognize driver’s gesture.However,the current work mainly relies on manual completion,which is not only inefficient,but also causes a waste of human resources.It is imperative to explore an accurate and fast automatic detection scheme of driver gestures.At present,in the field of video surveillance and behavior recognition,the research objects are mostly focused on the overall behavior of people,and the key parts of the action have not been finely split.Based on the actual operation of Beijing subway,this paper analyzes and discusses the following three gesture recognition methods by taking the gestures of drivers in the surveillance videos of drivers’ cabs as the research object:(1)Driver gesture recognition based on improved Kernel Correlation Filtering(KCF)algorithm.This method applies object tracking theory to gesture recognition and is based on the traditional kernel correlation filtering tracking algorithm.According to the characteristics of gestures,the algorithm was optimized and improved to track drivers’ hands and identify drivers’ gestures according to their motion trajectory and coordinates.The research results show that the recognition speed of the algorithm is good but the accuracy is not very high.(2)Driver gesture recognition based on Dense Trajectory and Fisher Vector coding techniques.Aiming at the problem of low recognition accuracy in the object tracking method mentioned above,this method takes a new approach to sample feature points of video clips containing driver gestures.After removing invalid sampling points,spatiotemporal features are extracted based on the trajectory formed by sampling points on continuous video frames.Gesture recognition is accomplished by means of mixed Gaussian model and Fischer coding technique.Experimental results show that this method improves the recognition accuracy greatly,but sacrifices the recognition speed.(3)Driver gesture recognition based on regional Convolutional 3D Network(RC3D).Aiming at the disadvantage that the above two methods cannot take into account the recognition accuracy and speed,and the research object is a simple video clip containing only a single gesture.From the perspective of deep learning,with the help of the feature extraction sub-network,the timing proposal sub-network and the behavior classification sub-network in the R-C3 D network,through the network training,optimization and adjustment,not only with higher identification accuracy to complete the driver gesture recognition,also can locate gestures at breakneck speed,the practical application of gesture recognition technology for the driver has important significance.Through the above research,the expected research results of this paper are as follows:for videos containing multiple gestures,the R-C3 D network model can not only realize gesture recognition well,but also accurately predict the starting time of each gesture.This research conclusion enriches the relevant theories in the field of video surveillance and behavior recognition of rail transit,and plays an important role in the realization of automatic gesture recognition and monitoring of urban rail train drivers in a practical sense.There are 55 pictures,13 tables and 83 references in the article.