一种改进的TLD目标跟踪算法
|
摘要
针对TLD目标跟踪算法在实际应用环境中其适应性和实时性较差的问题,提出一种基于ORB算法对TLD特征点跟踪模块改进算法。引入ORB特征点检测算法,使目标窗口中的特征点离散化,并在此基础上结合图像金字塔L-K光流法,减少光照对运动目标的影响,以此提高检测精度;在TLD的检测器中引入Kalman滤波器,对丢失或被遮挡的目标位置预测搜索,提高检测精度及速度。实验结果表明,改进算法在正确检测帧数小幅提高的情况下,检测帧率达到了每秒13帧左右,是原始算法跟踪速度的1.5倍左右,该方法可减少匹配特征点的数量,缩小检测区域,提高处理速度。
To solve the problem of poor adaptability and real-time performance of TLD(Tracking-Learning-Detecting)target tracking algorithm in practical application environment,an improved algorithm based on ORB for TLD feature point tracking module is proposed in this paper.By employing the ORB feature point detection algorithm,the feature points in the target window are discretized,besides that the image pyramid LK optical flow method to reduce the influence of illumination on the moving target,thereby improving the detection accuracy.In the TLD detector,the well-known Kalman filter is combined to predict the search for lost or occluded target positions,improving detection accuracy and speed.The experimental results show that the improved algorithm achieves a frame rate of about 13 frames per second with a small increase in the number of correctly detected frames,which is about 1.5 times the tracking speed of the original algorithm.This method can reduce the number of matching feature points and the detection areas,and then improve processing speed.
To solve the problem of poor adaptability and real-time performance of TLD(Tracking-Learning-Detecting)target tracking algorithm in practical application environment,an improved algorithm based on ORB for TLD feature point tracking module is proposed in this paper.By employing the ORB feature point detection algorithm,the feature points in the target window are discretized,besides that the image pyramid LK optical flow method to reduce the influence of illumination on the moving target,thereby improving the detection accuracy.In the TLD detector,the well-known Kalman filter is combined to predict the search for lost or occluded target positions,improving detection accuracy and speed.The experimental results show that the improved algorithm achieves a frame rate of about 13 frames per second with a small increase in the number of correctly detected frames,which is about 1.5 times the tracking speed of the original algorithm.This method can reduce the number of matching feature points and the detection areas,and then improve processing speed.
引文
[1] SMEULDERS A W M,CHU D M,CUCCHIARA R,et al.Visual tracking:an experimental survey[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2014,36(7):1442-1468.
[2] KRISTAN M,LEONARDIS A,MATAS J,et al.The visual object tracking VOT2016 challenge results[C]//Computer Vision-European Conference on Computer Vision 2016 Workshops,Amsterdam,2016:191-217.
[3] KALAL Z,MIKOLAJCZYK K,MATAS J.Tracking-Learning-Detection [J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2012,34(7):1409-1415.
[4] KALAL Z,MIKOLAJCZYK K,MATAS J.Forward-backward error:automatic detection of tracking failures[C]//International Conference on Pattern Recognition,Istanbul,Turkey,2010:2756-2759.
[5] KALAL Z,MIKOLAJCZYK K,MATAS J.Face-TLD:Tracking-Learning-Detection applied to faces[C]//IEEE International Conference on Image Processing,Hong Kong,China,2010:3789-3792.
[6] 秦飞,汪荣贵,梁启香,等.基于关键特征点的改进TLD目标跟踪算法研究[J].计算机工程与应用,2016,52(4):181-187.QIN F,WANG R G,LIANG Q X,et al.Improved TLD target tracking algorithm based on key feature points[J].Computer Engineering and Applications,2016,52(4):181-187.
[7] 祝贤坦,石繁槐.基于BRISK特征点改进的跟踪学习检测方法[J].计算机工程,2017,43(2):268-272.ZHU F T,SHI F H.Improved tracking learning detection method based on BRISK keypoints[J].Computer Engineering,2017,43(2):268-272.
[8] RUBLEE E,RABAUD V,KONOLIGE K,et al.ORB:An efficient alternative to SIFT or SURF[C]//International Conference on Computer Vision,Barcelona,Spain,2012:2564-2571.
[9] 李小红,谢成明,贾易臻,等.基于ORB特征的快速目标检测算法[J].电子测量与仪器学报,2013,27(5):455- 460.LI X H,XIE C M,JIA Y Z,et al.Rapid moving object detection algorithm based on ORB features [J].Journal of Electronic Measurement and Instrument,2013,27(5):455- 460.
[10] MUR-ARTAL R,MONTIEL J M M,TARD S J D.ORB-SLAM:A versatile and accurate monocular SLAM system [J].IEEE Transactions on Robotics,2017,31(5):1147-1163.
[11] WEBERRUSS J,KLEEMAN L,BOLAND D,et al.FPGA acceleration of multilevel ORB feature extraction for computer vision[C]//International Conference on Field Programmable Logic and Applications,Gent,Belgium,2017:1- 8.
[12] BOUGUET J Y.Pyramidal implementation of the Lucas Kanade feature tracker description of the algorithm[J].OpenCV Documentation,1999,22(2):363-381.
[13] KALAL,Z,MATAS,J,MIKOLAJCZYK,K.Online learning of robust object detectors during unstable tracking[C]//2009 IEEE 12th International Conference on Computer Vision Workshops,Kyoto,Japan,2009:1417-1424.
[14] 童源,费树岷,沈捷.基于TLD框架的快速目标跟踪方法[J].计算机应用研究,2018(1):317-320.TONG Y,FEI S M,SHEN J.Fast object tracking method based on TLD algorithm[J].Application Research of Computers,2018(1):317-320.
[15] KALMAN R E.A new approach to linear filtering and prediction problems[J].Journal of Basic Engineering Transactions,1960,82:35- 45.
[2] KRISTAN M,LEONARDIS A,MATAS J,et al.The visual object tracking VOT2016 challenge results[C]//Computer Vision-European Conference on Computer Vision 2016 Workshops,Amsterdam,2016:191-217.
[3] KALAL Z,MIKOLAJCZYK K,MATAS J.Tracking-Learning-Detection [J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2012,34(7):1409-1415.
[4] KALAL Z,MIKOLAJCZYK K,MATAS J.Forward-backward error:automatic detection of tracking failures[C]//International Conference on Pattern Recognition,Istanbul,Turkey,2010:2756-2759.
[5] KALAL Z,MIKOLAJCZYK K,MATAS J.Face-TLD:Tracking-Learning-Detection applied to faces[C]//IEEE International Conference on Image Processing,Hong Kong,China,2010:3789-3792.
[6] 秦飞,汪荣贵,梁启香,等.基于关键特征点的改进TLD目标跟踪算法研究[J].计算机工程与应用,2016,52(4):181-187.QIN F,WANG R G,LIANG Q X,et al.Improved TLD target tracking algorithm based on key feature points[J].Computer Engineering and Applications,2016,52(4):181-187.
[7] 祝贤坦,石繁槐.基于BRISK特征点改进的跟踪学习检测方法[J].计算机工程,2017,43(2):268-272.ZHU F T,SHI F H.Improved tracking learning detection method based on BRISK keypoints[J].Computer Engineering,2017,43(2):268-272.
[8] RUBLEE E,RABAUD V,KONOLIGE K,et al.ORB:An efficient alternative to SIFT or SURF[C]//International Conference on Computer Vision,Barcelona,Spain,2012:2564-2571.
[9] 李小红,谢成明,贾易臻,等.基于ORB特征的快速目标检测算法[J].电子测量与仪器学报,2013,27(5):455- 460.LI X H,XIE C M,JIA Y Z,et al.Rapid moving object detection algorithm based on ORB features [J].Journal of Electronic Measurement and Instrument,2013,27(5):455- 460.
[10] MUR-ARTAL R,MONTIEL J M M,TARD S J D.ORB-SLAM:A versatile and accurate monocular SLAM system [J].IEEE Transactions on Robotics,2017,31(5):1147-1163.
[11] WEBERRUSS J,KLEEMAN L,BOLAND D,et al.FPGA acceleration of multilevel ORB feature extraction for computer vision[C]//International Conference on Field Programmable Logic and Applications,Gent,Belgium,2017:1- 8.
[12] BOUGUET J Y.Pyramidal implementation of the Lucas Kanade feature tracker description of the algorithm[J].OpenCV Documentation,1999,22(2):363-381.
[13] KALAL,Z,MATAS,J,MIKOLAJCZYK,K.Online learning of robust object detectors during unstable tracking[C]//2009 IEEE 12th International Conference on Computer Vision Workshops,Kyoto,Japan,2009:1417-1424.
[14] 童源,费树岷,沈捷.基于TLD框架的快速目标跟踪方法[J].计算机应用研究,2018(1):317-320.TONG Y,FEI S M,SHEN J.Fast object tracking method based on TLD algorithm[J].Application Research of Computers,2018(1):317-320.
[15] KALMAN R E.A new approach to linear filtering and prediction problems[J].Journal of Basic Engineering Transactions,1960,82:35- 45.