Group Activities Recognition Method Based On Trajectory Energy Diffusion Map

The vision system is an important way for people to know and understand the world,the brain analyzes the external information obtained by the visual system and makes a corresponding response to the outside. Computer vision uses computers to imitate the human visual system to complete the acquisition, processing, storage and understanding of external information, achieving a variety of complex function. With the rapid development of computer hardware, computer development in terms of storage and computational performance change rapidly. The computer has successfully replace human to complete many complex work in daily life, behavior recognition is the successful application of computer in the video monitoring.The main content of this paper is the group action recognition research based on trajectory under a video sequence under. Its meaning is that extracting the effective information of people under different motion states, then getting features of trajectory through relevant algorithms, identifying different behaviors through analyzing these characteristics. The significant progress of target detection and tracking technology has provided support for group behavior recognition based on the trajectory. The behavior recognition has a wide application in video surveillance, virtual reality and motion analysis and so on, so it has attracted more and more attention of computer vision researchers.Group behavior recognition determines the behavior of moving target through analyzing the group behavior trajectory. Most of the existing methods in extracting characteristics of group activities cannot keep time information of trajectory suitably; moreover the movement of people belonging to the non rigid motion, including appearance, shape, exercise habits and other motion uncertainties. The noise of uncertainties will affect the extraction of trajectory characteristics. In order to solve the above problems, this paper proposes a group activities recognition method based on trajectory energy diffusion map. The method provides the trajectory energy block map which can perform and save time information of trajectoryeffectively. Along the direction of the trajectory, the color change of trajectory energy block reflects the time information of the trajectory. At the same time, the trajectory energy diffusion map can be obtained through processing the trajectory energy block diagram by using the energy diffusion principle. The trajectory energy diffusion map can smooth the noise of motion uncertainties. The recognition results can obtain higher accuracy. The using of feature fusion algorithm based on canonical correlation analysis for multiple features fusion ensures the diversity of characteristic information, thus improving the identification accuracy. The experimental results show that the recognition accuracy of the proposed algorithm is between 87% and 100% for the 7 sets of group activities of 567 video sequences.