Research on Monocular Visual Gesture Positioning based on YOLO v2

In order to ensure the speed and accuracy of positioning gestures during the human-computer interaction, YOLO v2 network with the best detection results in the target detection area was adopted in 2016 to generate the target detection data sets of the target areas with obviously different gestures. The YOLO v2 network is improved by the clustering of target box latitude, the pre-training of classification network, the screening training of multi-scale detection and the screening rules of the change candidate box to make it better adapt to the positioning task. It can locate the shooting gesture images acquired by the monocular camera in real time to the target area. The results show that the proposed method is effective and the average accuracy of the detection network in the target detection tasks of the captured images is increased to 79.5%.