Research On Video-based Limb Localization And Gesture Action Recognition Technology For Train Drivers

With the rapid development of my country’s railway transportation,the issue of railway safety has been paid more and more attention,which also puts forward a higher test for the guarantee technology of railway safety driving.According to statistics,one of the reasons for railway accidents is caused by the driver’s wrong operation.Therefore,this paper takes the subway driver as an example,starts from the gesture action detection of the subway driver,and designs and implements a system that can monitor and analyze the gesture action of the subway driver in real time by classifying and recognizing the gesture action of the subway driver as the research object.Do the following:1.Collect and produce datasets.Since there is no public data set of gestures of subway drivers,and there are few abnormal behaviors in the existing data sets,this paper records the imitation of the driver’s gesture behavior in the laboratory simulation environment.To test the robustness of the model,a small number of videos of partial gestures from the public datasets UCF101 and HMDB51 are also added to the dataset.2.Propose a multi-space-time scale fusion network RepC3D（Re-parameterization Convolutional 3D）.The target action area is obtained by the background subtraction method,and then the data set of the target area is converted into RGB and optical flow,sent into the network,and the video time series features and spatial features are extracted,and then the acquired features are processed through the RepC3D block.Logical operation to obtain fusion information,and finally through experimental comparison and experimental analysis,the average image recognition accuracy rate is 32% higher than that of the C3D model,and12.5% higher than that of the Rep VGG model.%,which is 12.5% lower than the Rep VGG model,and also has good performance in public datasets.3.Aiming at the research on the RepC3D network model,a gesture action monitoring system for subway drivers is proposed and designed.The system can effectively identify gesture behaviors by detecting existing and simulated data sets,and has certain value in practical applications.