A novel texture-based handcraft feature: spatial aggregated LD2oDP-VLAD for gait-based person recognition
Year: 2025
Type: article
Abstract: Gait recognition is becoming a biometric approach due to its passive acquisition and applicability in security and surveillance scenarios. However, existing handcrafted descriptors often fail to capture fine-grained variations and are sensitive to noise, limiting recognition performance. This work proposes a novel Local Directional Second-Order Derivative Pattern (LD2oDP) descriptor to encode sharper local intensity variations and preserve structural edge information. The extracted local features are aggregated using a Spatial Pyramid VLAD (SP-VLAD) framework to form a discriminative global representation. Experiments on the CASIA-B and CASIA-C datasets demonstrate improved recognition accuracy under viewpoint, clothing, carrying, and walking-speed variations.
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4415530484
