基于改进CStrack关联策略的多目标跟踪算法

Multi-object Tracking Algorithm Based on Improved CStrack Association Strategy

针对复杂场景下目标外观变化明显、运动不规律易导致轨迹中断和身份切换频繁等问题,从重识别(Re-Identification,Re-ID)特征、数据关联和插值等方面对跟踪器进行改进,提出基于改进CStrack关联策略的多目标跟踪算法。使用外观特征更新模块,减小因视角改变、目标移动导致特征剧烈变化而产生的影响,增强特征间的关联。提出二次关联方法,根据高低置信度检测结果的特点,使用不同的度量方式进行二次关联:第一次关联使用IoU距离融合外观特征作为关联的代价矩阵,第二次使用扩展IoU关联,缓解运动估计偏差、外观不可区分导致度量失效的问题;采用高斯回归算法,考虑运动信息,通过插值补偿漏检。在MOT17、MOT20数据集上进行测试,跟踪精度分别达到73.9%、64.2%。实验结果表明,该方法在跟踪精度上有明显优势,能够较好地适应复杂场景。

To address the problems of obvious changes in object appearance,irregular movements that lead to trajectory interruptions and frequent identity switching in complex scenes,the tracker is improved in terms of Re-Identification(Re-ID)features,data association and interpolation,and a multi-object tracking algorithm based on an improved CStrack association strategy is proposed.Firstly,an appearance feature update module is used to reduce the impact of drastic feature changes due to viewpoint changes and object movement,and to enhance the correlation between the features.Secondly,a two-stage association method is proposed,which uses different metrics for secondary association according to the characteristics of high-and low-confidence detection results;in the first association,IoU distance fusion appearance features are used as the cost matrix of matching,and in the second one,extended IoU association is used to alleviate the problems of biased object motion estimation and indistinguishable appearance leading to metric failure;the Gaussian regression algorithm is used to consider motion information to compensate for the missed detection by interpolation.Finally,the test on MOT17 and MOT20 datasets shows that the tracking accuracy is 73.9%and 64.2%,respectively,which indicates that the method has obvious advantages in tracking accuracy and can better adapt to complex scenes.