Human gait recognition based on X-T plane energy images

In this paper, we propose a novel algorithm for gait recognition. Binarized silhouette of a motion object is first segmented from color image, and then, spatio-temporal (XYT) volume is constructed by using these binarized silhouettes, and cut at knee and hip height. Next, energy images are extracted by projecting these three individual XYT volumes onto X-T plane, respectively. Fourier transform is employed as a processing step to achieve translation invariant for the silhouette sequences which are captured from the subjects walk in different speed. Then three frequency-domain feature vectors are fused. AdaBoost is used to select a small set of critical features from all of the features. Nearest neighbor and support vector machine (SVM) classifier are finally executed to produce final decision, respectively. The experiments are carried on one of the largest public gait database: the CASIA database. The experimental results show that the proposed algorithm is efficient for human gait recognition, and achieves competitive performance.