Moving Target Detection And Tracking Based On Random Ferns And Lucas-Kanade Optical Flow

Moving targets tracking and detection is the key issue in intelligent video surveillance,it is also the research foundation of subsequent target identification and behavior analysis.Moving targets tracking and detection is one of the hot topics in the field of computer vision,the technique has gained extensive application in many fields such as face recognition,behavior analysis,robotics,intelligent transportation and so on.Through these years,great achievements have been made both at home and abroad in targets tracking and detection,however,there are still many challenges such as deformation,illumination variation,occlusion,motion blur,scale variation and background clutters.In order to solve the problem of scale change and occlusion,this thesis made a deep research on the tracking and detection based on random ferns and Lucas-kanade optical flow.The feasibility of the proposed method was verified by simulation results,the system of target detection and tracking was designed and implemented.Main work of this thesis as follows:1.This thesis analyzed the current status of target tracking algorithm,summarized the technical difficulties and the disadvantages of the existing tracking methods,studied the KCF algorithm and proposed an improved method to solve the problem of scale change and occlusion.2.This thesis presented a method of object tracking based on multi-feature integration,HOG feature and CN feature were extracted to train the displacement classifier,HOG feature was used to analyze the gradient information of the image,CN feature focused on the representation of color information,and the final response fusion as the target position.Based on the high efficiency of multi-channel data,the anti-deformation ability of the algorithm is improved by the fusion of HOG feature and CN feature.3.This thesis presented a method of scale estimation based on Lucas-Kanade optical flow,the main idea of scale estimation is that the location of key points in the current frame are determined by using optical flow method,use the response map from the key points to assign weight to the distance ratio between matched points,and calculate the average distance ratio to estimate the scale variation.The proposed algorithm improved the accuracy of tracking.4.This thesis presented a method of online detection based on random ferns,a robust tracking algorithm requires a re-detection module in the case of tracking failure.an online detector based on random ferns was trained to re-detect,when the tracking response of the correlation filter is lower than a threshold value,the detector was used to detect the target and the result was used to update the target model.At last,a conclusion was made on the research of this thesis,the prospect which laid the foundation for the following research work was made on the further plan.