Research On Moving Object Detection And Classification Algorithm Based On Codebook Model

Intelligent video surveillance system in many ways to play a huge social and economic benefits,moving target detection and classification algorithms are the core technology of intelligent video surveillance system.In this paper,people and vehicles in the video as the main research object,combined with the actual application scenarios,focusing on the video moving object detection and classification algorithms are studied.In the aspect of moving object detection,aiming at the poor adaptability of the traditional algorithm in the dynamic background environment,it is easy to detect the changeable background as the moving object and the light is too sensitive to be prone to false detection.In this paper,a new dynamic codebook model target detection algorithm with cache update function is proposed.The algorithm can update the background model simultaneously in the process of detection and eliminate the influence of background changes.To ensure a more accurate test result and a more complete target contour,the improved Codebook model detection algorithm combined with five-frame difference method is used as the moving target detection method in this paper.Experiments show that this method can well adapt to the scenes of multi-moving targets,have strong adaptability to dynamic background environment,and can accurately detect moving targets in the scene of slow changes and abrupt changes of light.The target contour detected by the algorithm is clear and complete,which provides a guarantee for the target classification.In the research of classification of moving objects,through comparing and contrasting many kinds of classifiers,this paper uses the classifier based on Huff Forest algorithm to classify people and vehicles in the video.In order to improve the classification efficiency,it is optimized.In the process of feature extraction,a set of relatively stable and reliable subsets of static features is selected as the feature descriptor of the moving object.During the design of the classifier,the training phase and the detection phase are optimized respectively.The optimal segmentation threshold provides reliable parameters for the classifier to ensure the accuracy of classification.The classification speed of the algorithm is improved by randomly extracting the target image block.Experiments show that the optimized Huff Forest classifier can achieve a classification accuracy of 96% and a classification efficiency of nearly 60% higher than the original method.Thus,the target classification method has greatly improved the classification efficiency of moving targets on the basis of ensuring the accuracy of classification.