Research And Application On Active Shape Model

The location of moving objects has ever been hot topic in the field of computer vision research, and is widely used in intelligent monitoring, traffic control, medical diagnostics, human-computer interaction and other fields. Target tracking is divided into three key components: the detection of the target area, target tracking, target behavior analysis. In this paper, the detection of target area and target tracking are studied as focal points, the research methods are used in the video sequence analysis and related fields ,are provided theoretical support and application guidance for follow-up study.The main difficulties of Target tracking and location includes: object extracted from complex context, multiple objectives associated with and approximate, object partial occlusion. Academia and industry use some popular methods to solve specific problem, but it is difficult to solve all the problems. The methods based on deformable models solve the problem that non-rigid objects'shape often change and the features can not be accurate extracted, which has a good robustness when applied to non-rigid objects tracking. Among them, the Active Shape Model(ASM) is a representative method, and it is able to adapt to the target's contour variation within a certain range while maintaining the diversity of feature, even if under the complex background. This paper introduces the basic principle of the ASM algorithm, and its application for target location in color video sequence. To combine with target detection as pretreatment, the whole method can estimates the target's size and direction of motion very well, and give the correct initial position. The main work of the dissertation is as follows:(1) First, amend the local texture model of ASM. The local texture model of traditional ASM only use the gray information of object points to search. Two objects of different colors can be easily distinguished in color image. While they need to use the shape of the object, texture and other information to be distinguished because of the same or similar gray value in gray scale image, which is increasing the complexity of algorithm .This dissertation proposes an improved local search strategy based on the RGB color space multi-channel, and proposes a local texture model and search strategy with the color information of target. According to human vision character, RGB color space is converted to HSV color space. Instead of the gray information, the hue information is used as the model parameter. Finally through comparison of test results, the method combining with the hue information shows a better performance of object location in speed and accuracy.(2) Secondly, use Kalman filter to detect the target area as location pretreatment, improves the initial position accuracy, and provides a good basis for following research. Design a tracking method combining ASM algorithm with Kalman filter for non-rigid objects. The experimental results verify the validity and robustness of improved method, and show that the method can effectively enhance the location accuracy by single ASM algorithm for non-rigid objects which have obvious change in the shape scale and moving velocity, and show a good validity and robustness.