GRDDN: Enhanced Gait Recognition using a Deep Dense Network

Conventional recognition methods often fail to emphasize the reuse of features along with the separation of space-related and time-based characteristics regarding gait data. To mitigate these limitations, a unique approach is introduced called the Gait Recognition leveraging Deep Dense Network (GRDDN). Initially, continual gait sequence information undergo preprocessing steps, including cropping and normalization, before being passed through an elementary network unit to retrieve superficial gait features. These features are scrutinized further by the deep dense network, which integrates densely connected temporal-spatial feature separator unit and advanced convolutional block attention unit to extract separated temporal-spatial features effectively. The extracted gait features are subsequently split into probe and gallery features for facilitating accurate classification. Our method achieves recognition accuracies of 97.4%, 94.1 % and 80.6% during typical walking (NM), walking while carrying a bag (BG), and walking with a outerwear (CL) conditions respectively on the CASIA-B dataset.