Research On Objects Retrieval Method Under Mutiple Surveillance Cameras Based On Deformable Parts Model

Nowadays,image retrieval has more and more important research value under the background of massive data.For the intelligent video surveillance system,it has very important practical significance that how to retrieve specific target quickly and conveniently under multiple surveillance cameras.In this paper,we firstly carried out a systematic research and summary on image recognition and retrieval in recent years at home and abroad,especially the specific object retrieval methods under multiple surveillance cameras.Then we focused on the research of objects retrieval method under multiple surveillance cameras based on Deformable Parts Model(DPM).Aiming at the object retrieval in the surveillance video,the following tasks were completed:1.We studied and analyzed the present research on image and video retrieval methods and introduced the advantages and disadvantages of these different methods.Considering about the deficiency of existing methods,an object retrieval method under multiple surveillance cameras based on DPM model was proposed.2.The implementation principles,model training and target detection of DPM model were introduced in detail.Instead of current image content recognition method that all positive samples are labelled by people,the DPM model use the Latent Support Vector Machine(LSVM)for training which can save a large number of manpower.We did some model training about specific cartoon image and tuned parameters of the DPM model which were used to prove the validity of the DPM model.3.By using the DPM model,we designed the method of object retrieval under surveillance videos.The method used some process which are consist of pretreatment of videos,decoding frames,object detection,extraction for objects and objects matching to get the retrieval information of the same target in different videos.The DPM model is applied to practical application,and the validity and reliability of objects retrieval method under multiple surveillance cameras based on DPM model are further verified.