Research On Pedestrian Detection And Tracking In Intelligent Video Surveillance System

Intelligent video surveillance detects object such as pedestrians,vehicles in the acquisition of video surveillance data.The detection method is by using some target motion features or appearance features such as color,texture etc.combined with the detection window.The detection window can be a region of interest or saliency based,and also we can traversal each image using the sliding window.Target detection and tracking is the basis of action recognition,behavior analysis.However,the difficulty of detecting and tracking is greatly increased not only because of the constantly change of the background and light but also the effect of noise and target occlusion between each other.At the same time,the pedestrian detection method is trianed with some classifier models on the basis of the public data sets,but the detection rate will be greatly reduced when it is used to some other scenarios.This paper proposes a pedestrian detection framework based on transfer learning and sparse encoding,this framework can transfer an origin pedestrian detector trianed based on origin dataset to a new scene.The framework first gets the initial detect results by employing the origin detector to the new scene and then get some target templates that are labeled correctly by using some cues.In addition,it weights each target sample by using sparse coding combine with the similarities between target samples and target templates.At the same time,it gets the salient regions of origin samples and target templates by saliency detection method,and weights each origin sample by using sparse coding combine with the similarities between salient regions of origin samples and target templates.Finally,it retrians the detector combined with all weighted target samples and origin samples.The final detect rate based on our transfer learning framework increase 30%.This paper also presents a pedestrian tracking framework based on particle filter,we elaborates the principle of the particle filter framework,the proposed method can be used to track the pedestrian in the new scene,and the experimental results show that the proposed method is more robust than the traditional tracking method.