| Detecting and tracking moving human target is an important subject in visual analysis of human motion; it's also a valuable field in computer vision. This paper discusses detection and tracking of moving human target from static background in video sequences.Firstly, we use Adaptive Gaussian Background Modeling to segregate foreground pixels from static background. To be more responsive to the changing background, specifically to solve the "Ghost" issue of Adaptive Background Modeling, we provide algorithmic improvement and simplification, which, by using temporal learning rate of weight and parameter, increases the contribution of certain initial frames in the modeling. By comparing the results, the improved algorithm is apparently faster in eliminating "Ghost".Secondly, by aggregating foreground pixels, the target foreground objects can be segregated completely, based on which we can gather the characteristics of each foreground object respectively. Based on spatial continuity of target objects, with mathematical morphology and binary connected components analysis, we can reduce noise and smooth out gaps. By analyzing characteristics of capturing approximate background image with Single Gaussian Modeling, without losing the objective of reducing noise and keeping target objects complete, the threshold of segregating foreground pixels is determined by analyzing the result of subtracting background, which leads to definitive binary image of target object and background.As conclusion, starting with detecting and segregating target objects, based on characteristics in size, position and color of human target, we provide a moving object tracking algorithm, in which we use outside rectangle of moving objects to track human target in sight. When the targets are merging or parting, matching color template is used to keep the object's identity, which results in a good solution to the tracking issue when target objects are merging, disappearing, parting, or being shadowed. |
PDF Full Text Request