Research On Object Tracking Technology In Sequence Image

Object tracking technology is a key technique in the field of computer vision, ithas a broad range of application, especially in the field of video surveillance. Manydomestic and foreign researchers have conducted a thorough research and haveobtained several achievements. However, Object tracking under the complexenvironment still faces many problems, such as illumination change, backgroundinterference, occlusion and other factors which have a serious impact on the accuracyand stability of the tracking system. Aiming at the complex background, especially inthe case of illumination change, this paper conduct the research, the main work anddetails are as follows:1. Analyzed and compared several typical tracking algorithm, including thosebased on probabilistic forecasts, template matching and classification. This paper putforward the particle filter algorithm as the overall scheme of object tracking.2. In view of the traditional color histogram is susceptible to illuminationchange, this paper put forward the fuzzy color histogram as observation model inthe process of object tracking and adopted the fuzzy C-average clustering algorithm tocalculate. Fuzzy color histogram makes each pixel not only belong to a histogram bin,but also have a certain probability with all the bins, which greatly improves thestability of color features respect to illumination or noise.3. A new observation model combining fuzzy color histogram with gradientdirection histogram was put forward, considering the fact that observation modelbased on a single feature is not able to characterize the target under differentcircumstances accurately. According to distinction between foreground andbackground and the Fisher criterion, the weight of each feature can be determineddynamically. So the features can be optimally combined to characterize the targetmore accurately and distinguish the target from the background.4. A multiplicative illumination model and an illumination compensation modelwas presented in order to tackle the problem that feature varies under illuminationchange. The compensation model is a linear combination of the Legendre polynomial.When illumination changes, the model will be compensated, which makes the grayscale of the image similar before and after, and the features can be extracted forcomparison.5. Experiments have been conducted to verify two proposed processingscheme--feature fusion and illumination compensation.Experiments show that the proposed object tracking scheme can accurately andsteadily track moving object under illumination changes, which meets the trackingrequirements and achieves the expected effect.