Study On Object Tracking And Recognition Algorithm In Human-Robot Interaction

Higher requirements are put forward for Human-Robot Interaction (HRI) along with the rapid development of service robots. With the development of computer technology and robotics, HRI has gradually developed into an independent field of study. Vision based HRI is a technology that human beings interact with robots by means of vision sensor, which consists of human perception,human gesture recognition , social interaction and any other sub-topics. In this paper, human perception of vision based HRI was chosen as the point of penetration, and my study focuses on visual object tracking technology with human body or human face as the main objects and human face detection and recognition technology.In this paper, a multi-feature fusion based particle filter algorithm for visual object tracking is proposed to improve the robustness of visual object tracking . An adaptive method of choosing object color histogram is employed to get accurate color model of the object, in which single part color histogram or multi-part color histogram is chose as the color histogram of the object according to the color distribution of the object. Meanwhile, Edge orientation histogram is also employed as another assistance feature to cope with the influence of illumination variation and background confusion. The color feature and the edge orientation feature are two independent tracking threads, and which are fused by means of multiplicative strategy under particle filter frame in this algorithm.The results of the experiment show that the proposed algorithm can track both single object and multi-object even disturbed by the one with similar color.In this paper, a skin color feature based face detection and tracking algorithm using particle filter is proposed for the nonlinear and non-Gaussian characteristics of the state model and observation model in the visual object tracking field . First of all, the skin color model and ellipse template are used to detect the face region, and then weighted color histogram is chosen as the color distribution model of the target. The weights of the particles are updated based on the Bhattacharyya distance which is employed to represent the similarity between the face region and the particle. The proposed algorithm can achieve human face detection and tracking automatically.Face recognition is a small training sets of high dimension problem, an algorithm based on two-dimensional Gabor wavelets and support vector machine is presented to cope with this problem which is encountered in human face recognition field.In this algorithm, human face images are firstly processed by Gabor filters to get Gabor features space, and the space is then downsampled to reduce the dimensionality. Principal component Analysis is also employed to reduce the dimensionality of the Gabor features space significantly. After that we choose the Gabor features vector as the input of support vector machine to train so as to get the classifier of human face. Final experiment results conducted on the ORL and Yale face database show that the proposed method has high recognition rate and good robustness.