Object Detection And Recognition Algorithm Research And Application In Intelligent Video Surveillance System

The technology of object detection and recognition based on video sequence images is an important part in the intelligent video surveillance system, which have used widely in a variety of important places such as shopping malls, parking areas, banks, exhibitions and museums. The main idea of this technology is to use the algorithm to detect interest objects and to identify its condition in the video image sequences, in order to analysis if there are unusual event happened in the monitoring scene, such as theft and loss events. The technology of object detection and recognition is a research puzzle in the practical application field of intelligent video surveillance systems, image registration, and human-computer interaction, the research of object detection and recognition has wide application prospects, and this paper will solve some key issues of the related fields.Based on the existed matching algorithms, this paper mainly studied the objects detection and recognition algorithm based on the feature extraction and feature matching of the SIFT algorithm, and conduct a detailed analysis to the SIFT algorithm from four aspects, which is build scale space, detect feature points of the scale space, generate the feature descriptor and measure the feature similarity. The scale invariant feature has very stable characteristics to the non-rigid objects, complex background by experimental validation, radiation transform and illumination alternate, which have been widely used in various fields of computer vision. In order to get a batter extreme selection effect, join three-layer Gaussian blurred images in each group of top level in the extreme point detection process. In view of the results exist of one-to-many matching, the combination of algorithm of the SIFT feature extraction and SVD matching has been proposed in this paper. To achieve the purpose of removed mismatch points maximally and located the detected target accurately, we also proposed a constraint algorithm based on iterative density, we use the search box to search the fig of matching results, and calculate the number of matching points which falls into the search box to constrain the match points. We use the iterative density constraint method to eliminate the mismatch points of the matching results of SIFT-SVD algorithm, which not only can get the higher matching accuracy, but also can identify the location of the detected target in the image.Theft detection technology can identify anomalies in the monitoring scene in real-time, and send out the corresponding message, such as the lost event. Currently, theft detection technology have lower adaptation to the illumination alternate, complex and dynamic background, obscured, camera transfer, the target is non-rigid objects and other confounding factors, the SIFT-SVD feature matching algorithm based on the iterative density constraints will be applied in the field of theft detection in the video surveillance in this article, and we give some innovative improvements in the stage of feature extraction of objects, which extract the feature information of detected object in the six surface, and remove the duplicate features to improve the accuracy of feature detection. We re-built feature library to the precious objects, fix their original location and limit the location area in the scene, we use the feature information of library to match with the feature of the restricted areas in the video image sequence in the monitoring process, and analyze the matching results with the adaptive threshold algorithm, we think the object has been transferred when the results less than this threshold, and it will send out alarm signal automatically to remind the manager that an unusual event have happened. The technology can guard and ward the safety of the valuables against burglars in real-time, and protect the property safety. Object detection and recognition algorithms have been used to identify the thefts event in video surveillance in this paper, we detect multiple objects at the same time in the experimental, the algorithm proposed in this paper have been verified that it can identify whether object have been moved or stolen event accurately, and can identify different events occur simultaneously exactly, it have lower omission false rate, and higher detection accuracy which have achieved the expected demand.