Motion Detection And Tracking Based On Video

Motion detection and tracking based on video are computer vision's one of the fundamental problems and researching hotspots currently,which is widely used in industrial manufacture,mining,public security,road transportation,city management,building safety and so on.Especially in the oil industry,safety production problem concerns economic efficiency and personal safety.The application of motion detection and tracking can transform passive monitoring to active monitoring and accomplish the safety production's goal,including prevention before the event,treatment in the event,and analysis after the event.The key goal of motion detection and tracking is to handle video to get the region of variation,record the motion's trail,and analyze its behavior.Nevertheless,the actual scenarios are extremely complicated,which contain illumination changes,camouflaged foreground objects,and dynamic background.These challenges make motion detection and tracking extremely difficult.Especially in oil filed scenario,the circumstance could be more severe.To achieve better results,this paper is based on oversea and domestic research findings and carries out the following research.(1)Summary and analyze oversea and domestic research status on motion detection.This paper divides background subtraction motion detection into statistic,subspace learning,codebook,background sample set,mixed models.What's more,the paper analyzes these models' theories,detection processes,and improvements.In the meantime,the paper compares their advantages and disadvantages and selects background sample set as the paper's main framework.(2)Based on VIsual Background Extractor,coined ViBe algorithm,this paper proposes an improved perception-inspired dynamic threshold and spatio-temporal features background sample set motion detection approach.The ViBe has the advantages of taking up less space in memory,fast detection speed,and high detection precision.However,the algorithm can't adapt to the dynamic environment,which could cause some missing detection and false detection.Moreover,the algorithm may cause the ghost problem.Our algorithm applies <r,g,I> and SILTP to represent background model;makes use of GMM to initialize background;utilizes perception-inspired confidence interval to segment motion objects.In this paper,we choose SABS standard dataset and other classical algorithms to evaluate our method,and the method can get the best detection precision.(3)Based on advanced Self-Balanced SENsitivity SEgmenter,coined SuBSENSE algorithm,this paper proposes an improved background's dynamic factor and spatio-temporal features background sample set motion detection approach.The SuBSENSE has the advantages of adapting to the dynamic environment and high detection precision.However,the algorithm could cause some false detection for low brightness distribution areas.Moreover,the algorithm may cause the ghost problem.Our algorithm applies <r,g,I> and ELBSP to represent background model;makes use of median approach to initialize background;utilizes improved dynamic factor to adapt to the dynamic environment.Similarly,we choose SABS standard dataset and other classical algorithms to evaluate our method,and the method can get the best detection precision.(4)Propose a pedestrians' foot-point and head-point recognition methods based on motion detection results.Due to the complication of the environment and the limitation of detection algorithm's performance,it is difficult to get integrate and accurate motion object.We need recognition approach to obtain object's accurate location.For foot-point,principal component analysis is employed in getting pedestrian's upright direction and mapping the whole object to this direction.In the meantime,color feature,area feature,and position feature are utilized for detecting head-point.These methods can accurately obtain pedestrians' 2-D locations in real scenarios.These 2-D locations can be the basis of obtaining actual locations.(5)Select an appropriate object tracking algorithm to adapt to dynamic objects.In the real surveillance scenarios,there are some dynamic changes for the tracked object's appearance and scale,which could cause the object's lost problem.By comparing the traditional and advanced tracking algorithms,we choose spatio-temporal context tracking approach.(6)Perform video surveillance missions for the oilfield enterprise's scenario.Ultimately,this paper combines with the above mentioned methods to solve surveillance missions,including intrusion detection,object tracking,and pedestrians' spatial locations recognition.The proposed algorithms in this paper can get the best detection precision for the standard dataset and a good effect in practical intelligent surveillance application.Furthermore,these can be not only applied to oilfield enterprise's intelligent video surveillance,but also many other scenarios,such as military,industry,biomedicine,intelligent transportation,human-computer interaction and image compression.