Study On Human Infrared Gait Recognition Technique Of Hybrid-Dimensional Features From Motion Images

With the development of the artificial intelligence and machine vision technology, the integration of human motion analysis and biometric recognition becomes a hot topic. Gait recognition, as the only remote biometric authentication technology, mainly analyzes the motion images including walking human and recognises the human identification by his walking style. Many kinds of techniques, such as computer vision, pattern recognition, video and image sequences processing, are combined to accomplish this goal, which key point is to find a reliable method for extracting and classificating the gait characteristics. Focusing on this topic, this thesis did some exploratory researches on detecting the moving body in infrared gait sequence, extracting and classificating the gait features.Firstly, the moving body detection method in infrared gait sequence was investigated. Images were extracted from the gait video flows collected by the infrared camera and pre-processed through binarization, morphological filtering and connected component analysis before generating the moving human silhouettes by boundary tracking algorithm. And then the gait cycle was estimated according to the periodic variety of human silhouette width signals and used as the basic unit for future gait classification and recognitionSecondly, the three-dimensional and two-dimensional expressions of the infrared gait features were designed. Joint angles were extracted by the model-based tracking method. The three-demensional human model was built and projected. The similarity of the models and image was measured using the pose evaluation function which included the boundary and region characteristic. A hierarchical search strategy was used to extract the lower body joint angles. Gait features were also described by peak values of Radon transform from human silhouettes.Ultimately, the thesis tried to fuse and classify the hybrid-dimensional gait features. The feature-level data fusion was used to combine these two features. The k-fold cross validation with support vector machine (SVM) and the K-means classification were both employed for classifying and recognizing the gait features. The results showed that①The performance of hybrid-dimensional feature recognition was better than that of single-dimentional feature recognition;②The performance of k-fold cross validation with SVM was superior to that of K-mean algorithm in this study. Researches and applications about human recognition based on infrared gait are still on the initial stage. The achievement of this thesis may exploit a new way in the development of infrared gait recognition.