The Research On Face Tracking Algorithm Under Occlusion

With the development of information processing technology, face tracking hasbecome a hot spot problem. It has important applications in areas such as personalidentification, intelligent human-machine interaction system, video surveillance and soon. Face tracking technology is relevant to subjects such as pattern recognition, digitalimage processing, multimedia and so on. Some challenges are needed to be solvedurgently, including non-regularity motion state, illumination changes, efficiency of thealgorithm and occlusion problem. Based on previous academic research, the paper ismainly focused on how to resolve the occlusion problem, the research work includes:.1. The research status of face tracking technology is introduced and thealgorithms that proposed in recent years are presented; The principle and steps of theCAMSHIFT algorithm is introduced, the advantages and disadvantages of it isanalyzed and some experiments are implemented; The occlusion problem is analyzedand some usual solutions are introduced.2. A face tracking algorithm combined with skin characteristics is proposed:Thecontour of the skin region on the target is extracted after initialized the tracking target;The occurance of occlusion is detected by the changes of tracking window; Afterocclusion has occured, all the skin regions in each frame are extracted, and thecandidate targets is selected by comparing the size and the ratio of width and height ofthe region, and the target is determined by the similarity of the contour; the trackingstate is recovered as soon as the target is found out.3. A face tracking algorithm combined with feature points is proposed:The featurepoints of the target are detected by SURF operator and saved as the character; Theoccurance of occlusion is detected by the changes of tracking window;After occlusionhas occured, all the faces in each frame is detected by the face detecting algorithm andthe feature points of each face is extracted, the feature points is matched by the nearestneighbor method which is based on the coordinates, the target is found out by thematched result; the tracking state is recovered as soon as the target is found out.4. Two experimental systems are designed to verify the performance of thealgorithms, the efficiency and limitation of them are analylized.