A novel gait recognition using SRML learning with AP clustering

Abstract

The human identity and gender recognition from gait sequences with random walking Directions. First collect a new gait dataset, where people walk freely in the scene, and the walking directions are arbitrary and time-varying throughout the sequence. Some frames of a gait sequence from our dataset, as well as the segmented and aligned human shadow. The latest approaches make the idealistic statement that persons walk along a fixed direction. First preprocessing the input video and by background calculation find the object detection and cluster them into several clusters. For each cluster, compute the cluster-based averaged gait image as features. Then, propose a sparse reconstruction based metric learning method to classify the video and identify the gender and person and maximize the inter-class sparse reconstruction errors and minimize the intra-class sparse reconstruction errors. The discriminative information can be demoralized for human identity and gender recognition.