Research Of Moving Target Trace And Behavior Analysis Under Complex Background

Research of human behavior analysis involves video image processing, pattern recognition, computer vision and other research fields, which plays a pivotal role in smart surveillance, sports competition, military security and many other aspects. Accurate and rapid target detection and tracking is the premise and foundation for the subsequent research carrying out smoothly. The related algorithms are proposed higher requirements for the background is complex and varied under realistic conditions. Therefore, it is a great theoretical value and application prospects for target detection, tracking and behavior analysis under complex background.This paper, taking the video stream as the data source, proposes disposal and solving ideas for the common complex scenarios in practical environment, there are mainly following three aspects of the main research work:1. For the target detection in low-light conditions, this paper combines a variety of light compensation ways to gain the reinforcing effects with a certain adaptive capacity towards the light changes, makes the target effective information more prominent, and safeguards the smooth development of the follow-up study.2. Use Mean Shift algorithm for target tracking, take color feature in YCb Cr color space as the description of the target model, and employ weighted target and updated templates to adapt to the cumulative environmental noise and improve target discrimination during the iterative process, thereby reducing the iterative times and realizing the real-time targets tracking.3. The parameterized skeleton model represents various movements and gestures during human body movement process, transforming the image analysis problem into a linear algebra problems, greatly improving the speed of modeling; at the same time, calculates the partial offset anomaly membership combining fuzzy theory to indirectly describe overall abnormal behavior, greatly simplifying the analysis process, achieving quantitative analysis and qualitative judgment of the abnormal behavior.Experiments shows that the algorithms have good robustness, low time complexity and other advantages, can accurately detect abnormal human behavior.