Deep supervised hashing for gait retrieval.

Area of Science:

• Computer Science
• Biometrics
• Machine Learning

Background:

• Gait recognition is a key biometric for surveillance due to its non-intrusive nature.
• The increasing use of gait biometrics necessitates efficient large-scale data retrieval methods.
• The gait retrieval problem focuses on identifying subjects with similar gaits within extensive datasets.

Purpose of the Study:

• To develop an effective model for gait retrieval in large-scale datasets.
• To combine deep learning for gait feature extraction with hashing for efficient storage and search.
• To address the need for rapid identification in gait-based surveillance systems.

Main Methods:

• Proposed the Deep Gait Retrieval Hashing (DGRH) model, a supervised hashing method utilizing a deep convolutional neural network (CNN).
• Employed CNNs to capture semantic gait features and learn compact binary hash codes.
• Integrated a classification loss for similarity preservation and a quantization loss for hash code quality control.

Main Results:

• The DGRH model demonstrated promising results on benchmark datasets including CASIA-B, OUISIR-LP, and OUISIR-MVLP.
• The model successfully performed gait retrieval tasks, indicating its effectiveness.
• Achieved efficient storage and speed for gait retrieval operations.

Conclusions:

• The end-to-end deep supervised hashing model learns discriminative gait features effectively.
• The DGRH model offers efficiency in storage memory and retrieval speed.
• This approach is suitable for large-scale gait retrieval applications in surveillance and beyond.