Real-Time Movement Detection and Object Tracking in Video Sequences
|Course||Control Theory and Control Engineering|
|Keywords||Motion detection Multi- Gaussian background model Motion Tracking Mean Shift Theory|
Digital image processing and computer vision to the very rapid development of the research direction in recent years, is now widely used in military and civilian fields the intelligent machines access to external information and understanding of the world. Motion detection and target tracking are two of the most important applications of computer vision is the research content. The background of this research is to provide key information for autonomous mobile robots in indoor environments world modeling and path planning, navigation and other high-level decision-making, especially for the robot's environment monitoring, target following, obstacle avoidance tasks such as judgment and decision-making basis. Motion detection and target tracking as two relatively independent computer vision applications, text the algorithm theoretical studies and experimental verification. Motion detection using background subtraction algorithm based multi-Gaussian background model as the core algorithm, while the introduction of a Gaussian filter preprocessing and morphological processing algorithms as an auxiliary, a complete set of motion detection algorithms and experiment verification. In this paper, the defects of the algorithm itself lead to false detection of periodic large area, and put forward a new model update algorithm to be addressed; unable to overcome the lack of camera movement, shadow interference were also put forward a model reconstruction algorithm and shadow removal algorithm based on HSV space, and experimental verification of the validity and advantage of the improved algorithm. In this paper, based on the Mean Shift tracking algorithm as the core algorithm for target tracking,. The proposed method based on two-dimensional histogram feature and built according to the target color and grayscale information the target templates; Bhattacharyya coefficient to measure the similarity of template matching, combined with theoretical analysis and experimental results demonstrate its feasibility. Shortcomings of the algorithm, the result of intense scenes of the target feature is not obvious target motion tracking failure problems, the paper proposes an improved similarity evaluation method and the introduction of the Kalman predictor of improved methods and to compare validation experiments. The article also discussed the contact of motion detection and target tracking algorithm, to do a preliminary exploration work for cross-integration of multi-tasking computer vision. State of motion of the target area; motion analysis method based on motion detection results can be estimated by comparing the test results of the historical movement as well as the target area of ??the tracking algorithm based on motion detection results automatically, thus the original semi-automatic target tracking algorithm improved for fully automatic algorithm.