Research On Face Tracking Algorithm Under Complex Environment

Tracking algorithm is more stable than human in monitoring system acting as an"observer" role. The research on tracing algorithm has always been a hot point in the visualfield of object tracking because it is stability, efficiency, the economic as well as valuable.In practical application, tracing algorithm is short of broad usability due to the diversityand complexity of the situation. To maintain the effect, tradition algorithm should beimproved. Choosing the lower complexity algorithm can achieve the goal of saving computerresources.This paper improved traditional Camshift algorithm which tracks objects depend on thecolor of objects only. The Camshift algorithm is improved to search the object by using thematrix of hue and magnitude of gradient’s backproject to enhance the robustness.And in some cases, for example, when the object is covered, tracking objects are missing,we can predict the centroid of objects in next frame by fitting the trajectory of object. We caninitialize LK optical flow to track the objects by establishing the facial model of SURFfeature points according to the particular area faces SURF features.The improved LK optical flow tracking is more time-consuming than the improvedCamshift algorithm. When object moves out of the covered region, it is unnecessary to trackthe object using the improved LK optical flow. So a dynamic switching mechanism isdesigned to adapt the changing situation as well as save system resource. When the numberof SURF points is over threshold, algorithm initializes the improved Camshift algorithm,saving computer resources of tracking system.The experiments show that: the improved Camshift algorithm is more robust than thetradition, and it also can track the object immediately. Sparse optical flow tracking by SURFfeatures is more robust than traditional LK optical algorithm when the key points of facesdisappear causing of rotating of faces. And the algorithm in this paper can chose a suitablemethod for tracking according to the environment where the object is, reaching the aim thatalgorithm can track the object consequently.