Train Driver Behavior Detection Based On Ensemble Learning

From the train monitoring video,it can automatically,accurately and quickly identify and locate the body movements of the train driver.It has developed into an urgent requirement for the relevant national administrative departments to strictly regulate the driver’s manipulation behavior and ensure the safety of train operation.However,in the actual monitored train video,due to the complex background,blurred driver motion,and sparse effective frames of the monitored video,the problem of train driver body motion detection has become an extremely important and challenging computer vision technology research topic.In view of the difficulties and problems faced by the existing target detection algorithms at home and abroad when they are directly applied to the body motion detection of railway train drivers,this paper studies and proposes an improved frameby-frame analysis method to detect the body motion of railway train drivers..First,this detection method is based on the traditional 2D convolutional network technology,and proposes a body motion detector based on static video frames to detect various body motions of the driver;finally,the frame-by-frame detection method is improved,and the speed can be doubled.The fast-watch network for browsing videos solves the problem of a large number of invalid background frames in train monitoring videos,and improves the speed of detecting monitoring videos.In the task of detecting the body motion of train drivers based on surveillance video,quick-view Network not only achieved good results,but also has a fast detection speed.However,the detector can not well realize the detection of the driver’s body in the video.The specific requirements of motion detection,especially when the driver is moving or body movements,because the spatiotemporal and spatial information of the body movements in the video is not fully utilized,some video frames often appear in some driver’s body movements misdetection and Missing detection problem.To solve this problem,a video behavior recognition method based on 3D convolutional network is introduced,and an integrated learning method is used to fuse the two networks.The fusion of 2D and 3D convolutional networks has further obtained higher detection performance.First of all,we use the fast-watch network to obtain the driver’s body motion detection results in the key frames;then,use the box regression network to establish a recommendation module to obtain the temporal and spatial positioning of the driver’s body motion,and output the corresponding video segment of the driver’s body motion;Then,the 3D convolutional network-based video behavior recognition technology is used to recognize the driver’s body movements.Finally,ensemble learning is used to fuse 2D and 3D convolutional networks to further improve the accuracy of the model.The detection results in actual test scenarios show that the method of fusion of 2D and 3D convolutional networks based on ensemble learning proposed in this paper can achieve accurate and efficient detection of train drivers’ body movements.In the task of detecting body movements of train drivers in train monitoring video,the ensemble learning method proposed in this paper can obtain better detection results while realtime detection.