CASIA-E: A Large Comprehensive Dataset for Gait Recognition.

Area of Science:

• Computer Vision
• Biometrics
• Pattern Recognition

Background:

• Gait recognition offers unique advantages for surveillance but is hindered by a lack of large-scale, real-world outdoor datasets.
• Existing datasets are often collected indoors, failing to capture the complexities of natural environments like dynamic backgrounds and varied lighting.

Purpose of the Study:

• To introduce CASIA-E, a novel, large-scale outdoor gait dataset designed to overcome the limitations of current resources.
• To provide a comprehensive benchmark for gait recognition research and development in practical scenarios.

Main Methods:

• Collected nearly one million gait videos from over one thousand subjects across diverse outdoor scenes and seasons.
• Recorded videos with 26 view angles, incorporating variations in clothing, bag carrying, and walking styles.
• Included soft biometric features (age, gender, height, weight, nationality) for each subject.

Main Results:

• The CASIA-E dataset contains a vast amount of video data with rich attributes and significant spatial-temporal variations.
• Experimental benchmarks were established, analyzing factors like large-scale training, vertical view angles, and different walking styles.
• The study explored the potential of thermal infrared gait recognition.

Conclusions:

• The CASIA-E dataset and its accompanying benchmark are expected to significantly advance gait recognition research.
• This resource will facilitate the development of more effective gait recognition systems for both academic and industrial applications, particularly in challenging outdoor environments.