Gesture Recognition of Subway Drivers Based on Improved Dense Trajectory Algorithm

Metro drivers need to make corresponding gestures when the train is in different operation states. The purpose is to confirm whether the display of each instrument on the console is normal, and then determine whether the train is in good running condition, whether the switch or signal on the line is working normally, and whether the door is closed normally. This information needs to be fed back to the train control center in time. Therefore, it is necessary to recognize the drivers' gestures in the surveillance video of the drivers' cab. This paper proposes a gesture recognition model based on the monitoring video of subway train cab, which combines Improved Dense Trajectory (IDT) algorithm, Fisher vector coding technology and Support Vector Machine (SVM) classification technology to realize feature extraction, feature coding and classification of video clips. The model recognizes the drivers' gestures in the actual train monitoring video clips, and the accuracy can reach 100%.