Currently,most of the skeleton-based gait recognition methods improve the accuracy of gait recognition by improv-ing the representation of spatiotemporal relationship,but ignore the linkage between the nodes of human physical structure and the cues of long and short time series.Based on the above problems,the spatial topological linkage of gait recognition is explored.In this paper,a new strategy method based on local partitioning is proposed to divide unstructured regions of the human body and con-struct the adjacency relationship between the regions to represent the relative changes of human posture.Secondly,a multi-scale temporal module is designed to extract multi-scale temporal features by associating node changes of local partitions with time as the axis.Through experimental verification,the proposed method surpasses the mainstream gait recognition methods on gait dataset CASIA-B,and the Rank-1 accuracy rate reaches 87.6%,77.6%and 72.8%respectively under different walking conditions.
维普
万方数据