Gait Feature Recognition Based On Energy Image Decomposition

With the fast development of information technology, information security and public safety has become a growing concern. And how to accurately and effectively identify one person's identity has become a research hotspot. Compared to the traditional method, such as ID cards, fingerprint, iris, handwriting and face recognition technology, etc., gait recognition has its own advantages. Gait recognition is a way to determine the identity of distance, so it has the feature of non-invasive, hard to hide, and low on system resolution requirements. As a new biometric, gait recognition has important research significance and wide application in the future. Current research on gait recognition is still in its infancy, researchers need to continue in-depth exploration and research. This paper carried out research in the following aspects:(1) Analysis on gait detection algorithm used. This paper selected a simple and fast targets detected method named background subtraction, using morphology and connected component analysis to reduce noise and holes. And then, use the periodic changes of aspect ratio of the human body to extract the periodic image sequences.(2) Based on the gait energy, this paper proposed a method of energy segmentation. Gait energy is divided into two parts. One part is the static gait energy images; the other part is the dynamic gait energy images. And then different moments were used to extract gait features from the static and dynamic gait energy images.(3) According to the idea of information fusion, this paper proposes an algorithm that combines static and dynamic gait features. Zernike moments and wavelet moments are used to extract gait features from the static and dynamic energy map. Using moments to extract the feature would cause too many dimensions vectors. Therefore, principal component analysis (PCA) is used to reduce the dimensions. Then use the method of information fusion to make the two features fused. So that, the static and dynamic features in the gait sequence images are effectively used.(4) Because of small number of samples in the gait database, using SVM method in the recognition process.(5) All the algorithms are experimented on the CASIA Gait Database. The database includes both normal walking, backpacks cloak and others walking state. The result shows that the algorithms achieved a good recognition rate in both small (20 samples) and large (124 samples) samples.