A Three-Way Auto-Correlation Based Approach to Human Identification by Gait

We propose a scheme for gait recognition using cubic higher-order local auto-correlation (CHLAC), discriminant analysis, and k-NN decision rules. CHLAC is based on three-way (x-, y-, and time-dimensional) auto-correlations of pixels in motion images, and it effectively extracts motion features. The method has several properties preferable for recognition: shift-invariance (rendering the method segmentation-free) and robustness to noise in data. Moreover, the method is so general as to use neither a priori knowledge nor heuristics about objects such as human shapes and is applicable to any three-way data. We made the scheme more effective for gait recognition by introducing some knowledge of gait to optimise parameters in CHLAC. Our scheme was applied to the NIST gait dataset for human identification, and the result was compared to those of other methods. Our scheme outperformed the others in spite of the simple feature extraction and the simple classification rule.