Dahua AI Gait Recognition Breaks CASIA-B Gait Records

The average accuracy of Rank1 for NM (normal walking), BG (with a bag) and CL (in a coat) reached 97.4 per cent, 94.0 per cent, and 87.0 per cent respectively, hitting another historical heights and maintaining its leading position. Gait recognition uses body shape and walking posture to identify a person, even if his/her face is occluded. It is one of the biometric recognition technologies with the greatest potential for long distance recognition scenarios. Aiming to address the technical difficulties of gait recognition in clothing changing search, carrying changing search and cross-view search, Dahua Technology integrates innovation and application of multi-modal gait algorithms, local gait feature extraction, and spatio-temporal gait feature extraction technologies. Combined with powerful model training and object recognition, it greatly improves the algorithm's robustness in special scene applications such as clothing changing, similar clothing, facial occlusion, and facial disguise, and thus making the gait recognition analysis more accurate and efficient.