Moving Object Detection Algorithm Based On Background Modeling In Surveillance Video

Video surveillance has been widely used in water transportation, the need especially for size, speed and flow of moving ships and analysis of abnormal events. Moving object detection, as the core of intelligent surveillance system, is the foundation of subsequent object tracking and behavior analysis. However, the actual surveillance environment is complicated and changeable, such as changing light, shaking leaves and Water ripples, which will influence the accuracy of moving object detection. It is necessary to study motion detection algorithms under complex environment. This thesis stems from the project named 《Research and Application of Key Technologies in Real-Time Video Monitoring and Recognizing System for Ships》 and 《Ship Ultralimit Detection System》, combines with the difficulties and challenges encountered in the project such as water ripples, stern drag lines, camera jitter and low speed of moving ships, analyzes overseas and domestic research actualities of background modeling algorithm, finds out the crux of error detection, proposes optimization scheme aiming at algorithm defects, and verifies the effectiveness of the algorithm through experiments.The thesis describes overseas and domestic research actualities of video surveillance system, analyzes the role and status of background modeling algorithms in moving object detection, compares the performance of classical background modeling algorithms, puts forward a new moving object detection algorithm based on modified VIBE algorithm, optimizes the MOG algorithm based on pixel motion feature, and presents an overrun monitoring algorithm of moving ships based on 3d image.The thesis studies the theories of background modeling, elaborates the advantages and disadvantages of six classical background modeling algorithms, verifies the detection results and applicable scenes by experiments.The thesis analyzes the strengths and weaknesses of VIBE algorithm, aiming at the problem of dynamic background, optimizes the algorithm in three phases including background model initialization, model matching and model updating. Background model initialization is conducted with several continuous frames in order to weaken the interference of "ghose". Matching threshold is self adapted according to background dynamics in order to reduce error detections caused by dynamic background. Discrete degree of background samples is calculated to find out the best sample needs to be replaced, conbining with spatial coherence and fuzzy theory, the accuracy of background model is improved and the false detection rate is reduced.The thesis elaborates the motion information distributions of background pixels under camera jitter, extract the pixel motion feature to secondly distinguish moving objects obtained from MOG algorithm, remove the error detections caused by camera jitter. Learning rate is also optimized, which is now self-adapted in foreground and background region. By this means, the contradiction of error detections and failed detections is resolved and the robustness of the algorithm is improved.The thesis studies the overrun monitoring alegorithm of moving ships based on 3d image, utilizes three scanners to pick up 3d point cloud, classifies datas by principal component analysis, and measures the 3d size of moving ships.The innovations of this thesis are as follows:· A moving object detection algorithm based on modified VIBE is proposed:Utilize several continuous frames to initial background model, adjust matching threshold automatically according to background dynamics, calculcate background samples’discrete degree to find out best replacement, conbines with fuzzy theory, the accuracy of background model is improved and false detections caused by dynamic background are reduced.· MOG algorithm is optimized by appling pixel motion feature:pixel motion feature is applied into MOG algorithm to remove false detections caused by camera jitter, and learning rate is adaptively adjusted to improve the robustness of the algorithm.· An overrun monitoring algorithm of moving ships based on 3d image is presented to measure 3d size of moving ships, which will provide a basis for monitoring the overrun of ships.