适用于电视交互的视觉手势识别算法设计

Design of Visual Gesture Recognition Algorithm for TV Interaction

介绍一种视觉手势识别算法在智能电视上的设计方案,用于在某些特定场景替代遥控器完成与电视的交互。该方案利用目标检测的方法,直接通过图像序列存在的手部检测和手势分类实现静态手势识别,通过连续多帧的位置关系判断来实现动态手势识别。算法设计采用YOLOv4-tiny一阶段检测模型架构,并针对移植终端的性能对识别算法进行改进,从而使手势操控在目标设备上完美达成。

This paper introduces a design scheme of visual gesture recognition algorithm on smart TV, which can replace the remote control to complete the interaction with TV in some specific scenes. Using the method of target detection, the static gesture recognition is realized by hand detection and gesture classification of image sequence, and the dynamic gesture recognition is realized by the judgment of position relation of continuous multiple frames. YOLOv4-tiny one-stage detection model architecture is adopted and the algorithm transplanted to the terminal is improved, so that all indicators of gesture control can be perfectly achieved on the target device.