Abstract:
Re-ID via gait features in 3D skeleton sequences is a new topic with many benefits. Solutions use hand-crafted descriptors or supervised gait representation learning. This paper proposes a self-supervised gait encoding method to learn person Re-ID gait representations from unlabeled skeleton data. We first create self-supervision by learning to reconstruct unlabeled skeleton sequences reversely, using richer high-level semantics to improve gait representations. Self-supervised learning is improved by exploring other pretext tasks. Second, inspired by the fact that motion’s continuity endows adjacent skeletons in one skeleton sequence and temporally consecutive skeleton sequences with higher correlations (referred to as locality in 3D skeleton data), we propose a locality-aware attention mechanism and contrastive learning scheme to preserve locality-awareness on intra- and inter-sequence levels during self-supervised learning. Finally, using context vectors learned by our locality-aware attention mechanism and contrastive learning scheme, Constrastive Attention-based Gait Encodings (CAGEs) represent gait effectively. Our method outperforms skeleton-based methods by 15-40% Rank-1 accuracy and many multi-modal methods with RGB or depth information.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
