A Lightweight Efficient Person Re-Identification Method Based on Multi-Attribute Feature Generation

Person re-identification (re-ID) technology has attracted extensive interests in critical applications of daily lives, such as autonomous surveillance systems and intelligent control. However, light-weight efficient person re-ID solutions are rare because the limited computing resources cannot guarantee accuracy efficiency detecting features, which inevitably results performance bottleneck real-time applications. Aiming at this research challenge, study developed a lightweight framework for generation multi-attribute feature. The mainly consists three sub-networks each conforming to convolutional neural network architecture: (1) accessory attribute (a-ANet) grasps ornament information an descriptor; (2) body (b-ANet) captures region structure (3) color (c-ANet) forms descriptor maintain consistency person(s). Inspired by human visual processing mechanism, these descriptors (each “descriptor” corresponds individual person) integrated via tree-based feature-selection method construct global “feature”, i.e., serving key identify person. Distance learning is then exploited measure similarity final re-identification. Experiments have been performed on four public datasets evaluate proposed framework: CUHK-01, CUHK-03, Market-1501, VIPeR. indicate that feature outperforms most existing feature-representation methods 5–10% rank@1 terms cumulative matching curve criterion; time required recognition low O(n)