Four-direction traffic police gesture recognition method based on time sequence linear human body skin model and graph convolutional network

• Main Authors: ZHANG CHENG, HE JIAN, LYU MENGFEI, XIONG ZHEBO
• Format: Patent
• Language: Chinese; English
• Published: 03.03.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention discloses a four-direction traffic police gesture recognition method based on a time sequence linear human body skin model and a graph convolutional network, and belongs to the field of electronic information. According to the method, a power tree of traffic police command gestures is reconstructed by using a sequential linear human body skin model (SMPL), and a traffic police gesture sequential dynamic graph model is constructed according to time context information. Secondly, for the problem that an existing graph structure division strategy is limited, a graph convolution kernel label division strategy (RHPS) based on the relative height is provided, and a graph convolution network (RHGCN) based on the relative height is designed; and finally, fusing the RHGCN and a spatial domain average predictor (SMP) design to realize a four-direction traffic police gesture recognizer MTPGR based on monocular vision. The task requirement of four-direction traffic police gesture recognition is effectively