目的 复杂环境下，运动目标在跟踪过程中受尺度变换以及遮挡因素的影响，跟踪准确率较低。针对这一问题，提出了一种遮挡判别下的多尺度相关滤波跟踪方法。方法 首先选取第一帧图像的前景区域，训练目标的位置、尺度滤波器和GMS检测器。然后，通过位置滤波器估计目标位置，尺度滤波器计算目标尺度，得到初选目标区域。最后，利用相关滤波响应情况对初选目标区域进行评估，通过相关滤波响应值的峰值和峰值波动情况判断是否满足遮挡和更新条件。若遮挡，启动检测器检测目标位置，检测到目标位置后，更新目标模型；若更新，则更新位置、尺度滤波器和GMS检测器，完成跟踪。结果 本文使用多尺度相关滤波方法作为算法的基本框架，对尺度变化目标跟踪具有较好的适应性。同时，利用目标模型更新机制和GMS检测器检索目标，有效的解决了遮挡情况下的目标丢失问题。在公开数据集上的测试结果表明，本文算法平均中心误差为5.58，平均跟踪准确率为94.2%，跟踪速度平均可达27.5帧/秒，与当前先进的跟踪算法相比，本文算法兼顾了跟踪速度和准确率，表现出更好的跟踪效果。结论 本文提出了一种新的遮挡判别下的多尺度相关滤波跟踪算法。实验结果表明，本文算法在不同的尺度变换及遮挡条件下能够快速准确跟踪目标，具有较好的跟踪准确率和鲁棒性。
Multi-scale correlation filter tracking algorithmbased on occlusion discrimination
Liu Wanjun,Zhang Zhuang,Jiang Wentao,Zhang Shengchong(Liaoning Technical University)
Objective Visual target tracking has become a research hotspot in the field of artificial intelligence at home and abroad, which widely used in national defense security, industry and people''s daily life, such as military recognition, security monitoring, pilotless automobile, human-computer interaction. Although great progress has been made in the past decade, the model-free tracking is still a tough problem due to illumination changes, geometric deformation, partial occlusion, fast motions and background clutters. The traditional methods of target tracking generally track the target by visual features. In case of the simple environment, these trackers can perform well for specific targets. Recently, visual object tracking has been widely applied to object tracking field due to its efficiency and robustness of correlation filter theory. A series of new advances of target tracking have been introduced and much attention has been achieved. A novel approach to predictive tracking, which is based on occlusion discriminant multi-scale correlation filter tracking algorithm,is proposed to overcome the problems of low accuracy caused by occlusion, scale changes in the tracking process under complex environment. Method On the basic framework of DSST (Discriminated Scale Space Tracker), a multi-scale correlation filter tracking algorithm is proposed. Reliability discrimination for the results of correlation filter response, which contribute to tracking stably for a long time, means to occlusion discrimination and update discrimination by the peak value and multiple peak fluctuation of the response map. The proposed algorithm in this paper can be summarized as two main points：1) Two kinds of calculation model were designed for the maximum peak and multiple peak fluctuation. By evaluating the tracking results according to above two models, we can determine the occlusion of the target and whether the target needed to be updated. 2) Redetect the missing target by the detector based on GMS (Grid based Motion Statistics). When the target is occluded, the GMS detector has been trained start to detect the target and locate the target again. Here''s the concrete tracking process: Firstly, the foreground area of the first frame image was selected, and train target position filter, scale filter and GMS detector. Then, the target location is estimated by the translation filter and the target scale is calculated by the scale filter. Performing correlation between the candidate samples obtained with different scales center on the new position and the scale correlation filter can derive the primary target area and the maximum response scale is the current frame image scale. Finally, the primary target area is evaluated by the correlation filter response, and the occlusion and update conditions are determined by the peak value and multiple peak fluctuation of the correlation filter response values. When both peak value and multiple peak fluctuation are mutated, it indicated that the target is occluded at the moment. The greater the mutation, the greater the degree of occlusion .In this case, update needs to be avoided to prevent tracking drift. If the target is occluded, the detector detects the target position and updates the target model after detecting the target location. When the peak value of the correlation filter response is greater than the historical value, and the peak fluctuation does not mutate, it indicated that the target information at this moment is more complete than that at time t-1, and correlation filter need to be updated .If the target needs to be updated, update location filter, scale filter, and GMS detector to complete tracking. Result The multi-scale correlation filtering method is used as the basic framework in our algorithm, and it has good adaptability to the target tracking of scale transformation. At the same time, using the target model updating mechanism and the GMS detector retrieving the target, solved the target loss problem in the occlusion effectively. This paper selected 9 challenging video sequences including Box, Bird1, Lemming, Panda, Basketball, DragonBaby, CarScale, Bird2, Girl2 from the public dataset OTB - 2013 and OTB - 2015 as well as a video data car_Xvid to conduct the experiments. The test results on the public data sets show that the algorithm in this paper has a lower average center error with 5.58, and has a better tracking accuracy with 0.942, and has a better tracking speed with 27.5 frames per second, compared with state of the art tracking algorithm DSST, KCF (Kernel Correlation Filter), LCT (Long-term Correlation Tracking), Staple, GOTUTN (Generic Object Tracking Using Regression Networks), FCNT(Fully Convolutional Networks Tracking), the algorithm show better tracking performance with higher tracking speed and tracking accuracy. Conclusion Based on DSST correlation filtering tracking, a multi-scale correlation filtering method based on occlusion discrimination is proposed. The experiment results show that the algorithm solved the problems of losing goals due to occlusion and error accumulation because of the continuously updated strategy effectively, and achieved stable tracking under the occlusion and multi-scale change. Compared with the current popular tracking algorithms, this algorithm has the following remarkable advantages: 1) Solved the problem of losing goals due to occlusion in DSST algorithm. 2) Detecting the occlusion and determining whether to update can mitigated the tracking drift problem in frame update. Doing like this not only can reduce the unnecessary update time, but also can significantly improve the tracking speed and tracking accuracy. This paper presents a new multi-scale correlation filtering tracking algorithm based on occlusion discrimination. Experiments show that the proposed algorithm can track the target quickly and accurately under the conditions of different scale transformation and occlusion, and has better tracking accuracy and robustness.