Occluded Object Detection Algorithm Based On Probability Density Distribution

Object detection has been research focus in the field of computer vision and pattern recognition,which is widely applied in the intelligent transportation,human-computer interaction,image analysis,electronic entertainment and other relevant aspects.In recent years,object detection has been more and more concerned which is related to many areas.Occlusion occurs frequently in daily life,which often brings besetting to people and causes inconvenience in life.Occlusion has been recognized as one of the most difficult areas within the object detection,and it has become the direction of efforts for scholars how to improve object detection rate with partial occlusion problem.Deformable part model which is widely applied has achieved a very good detection performance and been in the leading position,especially in the respects of body posture and gesture recognition.However,there are still some disadvantages for the deformable part model when it handles occlusion problem.When the detection window of the target object is partially obscured,such as overlaps between people and occlusion between people and object,it's obvious that the responses of part classifiers are dropping so that it can't be detected.In this thesis,the occluded object detection model based on probability density distribution is proposed,which takes advantage of the good characteristics of the deformable part model,to improve the performance of pedestrian detection with occlusion.Finally,this thesis shows the good experimental results in INRIA database,occluded INRIA database and ETHZ database.