Research On Moving Human Object Tracking Algorithms Based On Multiple Features

With the development of related technologies on computer vision and patternrecognition，human motion detection and tracking have attacked a lot of attention，and is usedwidely in some fields. Human tracking in real-world scenes is generally complicated，due tothe background changes，occurrence of occlusion，noise and other interference factors，so itbecomes hard to meet the needs of accuracy、robustness and real-time simultaneously. In thispaper，we design and implement a moving human tracking algorithm based on multiplefeatures，for the research of single and stationary camera。Compared with the tracking algorithm based on simple color feature，the proposedalgorithm uses both color and contour features to build model. It describes the object moreprecisely，and overcomes the sensitivity to lighting condition in environment. Combined withthe particle filter，the algorithm can adapt to the human body’s unlinear motion. Thereforewhen occlusion or overlapping happens，target’s state can be predicted，improving trackingrobustness. We adopt adaptive model updating strategy during tracking process，thus increasethe adaptability to light and object deformations. So accurate tracking can be achieved. Inorder to resolve the problems in manual selection，we also introduce temporal differencingand realize an auto tracking mode，extending the algorithm well. On the basis of these work，an auto face tracking scheme is implemented，and provides an efficient target tracking methodat close distance. Experimental results and analysis show that the algorithm not only ensuresthe accuracy，but also is real-time. It has good performance under some complicatedconditions.