Research And Implementation On The Related Image Processing Technologies Of Video Surveillance System

Intelligent Video Surveillance System (IVSS) is a combination of many objects, such as machine vision, image engineering, pattern recognition, artificial intelligence, computer science and so on. It is full of challenges and has a wide range of applications. It can automatically analyze the images captured from the camera, locate, track and recognize the objects in the monitoring scene, analyze and judge their behaviors, and then make an expected warning. The IVSS can assistance or even replace human to complete the task of monitoring or control. It has the advantages of initiative, round-the-clock, real-time, take measures before happen, etc. It not only reduce the burden of the staff, reduce the cost, but also increate the safety. With these advantages, the IVSS will be widely used in fire alarm, public security surveillance, military reconnaissance, crowd control and other fields.With a large numbers of researchers around the world devoting themselves to the technologies of the IVSS, intelligence video surveillance has been made a great progress. But there are still many problems need to be improve, in the field of theory and application. Base on the achievement of the previous, some related image processing technologies of the IVSS are studied. The main work can be summarized as follows:Firstly, object detection technologies are investigated. After analyzing and comparing some common methods of moving target detection, the method of background subtraction based on Gaussian mixture model was chosen. The process of building and updating the Gaussian mixture model was analyzed. The algorithm can effectively deal with the problems of building and updating the background model, background disturbance, light changes etc. Based on the RGB color model, a shadow detection algorithm was analyzed for detecting and eliminating the shadow areas. For solving the problem that the noise which could impact object feature extraction existing in the result of initial moving object detection, the morphological noise filter method was introduced.Secondly, the relate theories of image stitching technology was expounded. After analyzing the advantages and disadvantages of several image feature extraction methods, the SIFT feature extraction algorithm was introduced specifically. The SIFT algorithm has a property of invariance in rotation, scaling and translation. It analyzes the algorithm of BBF and RANSAC, which are used for matching the SIFT feature points, and the image transformation model, and then compares the result of several image fusion methods.Thirdly, the basic theory of particle filter,"principle of Bayesian filter"and"Monte Carlo method"was introduced. The principle of particle filter, the dynamic state model and observation model of objects in the particle filter was analyzed, and also introduces the process of realization the important sampling.Lastly, a SIFT based abnormal object detecting algorithm was proposed. The algorithm detects the abnormal object or changes in the surveillance scene by matching the camera image and the panoramic image. And a multi-face tracking algorithm base on particle filter was proposed, for the purpose of introducing the application of particle filter in multi-object tracking. The algorithm matches and tracks the faces. Based on proposed abnormal object detecting algorithms and the multi-object tracking algorithm, the working flow frame of an intelligent video surveillance system was proposed.The experimental results of the proposed algorithm show their effective, and also the realizability of the proposed intelligent video surveillance system.