基于改进CenterTrack的多目标跟踪算法

Multi-target Tracking Algorithm Based on Improved CenterTrack

相较于传统的两阶段多目标跟踪算法,基于锚点的多目标跟踪算法CenterTrack通过在单一网络下同时完成检测与跟踪,保证较高跟踪精确度,同时又具有更快的推理速度,适合行人多目标的实时跟踪,但面对行人数量密集或者遮挡严重情况难以准确检测行人位置。为解决该问题,提出基于改进CenterTrack的多目标跟踪算法,在骨干网络中加入注意力机制以提高特征图的检测精度;设计了强化检测的跟踪算法,完成多目标跟踪的同时增强下一帧的检测效果,从而提高跟踪的精确度;通过使用数据增强方法处理过的数据集进行训练,算法的泛化能力得到提高。在MOT17数据集上的验证实验结果表明:与原算法相比,本文改进算法具有更高的跟踪精确度和正确检测比。

Compared with the traditional two-stage multi-target tracking algorithm,CenterTrack,a multi-target tracking algorithm based on anchor-free,completes detection and tracking in a single network at the same time to ensure a high tracking accuracy and has a faster reasoning speed,which is suitable for multi-target real-time tracking of pedestrians.However,it is difficult to accurately detect the position of pedestrians of the densely populated place in the image.To solve this problem,a multi-target tracking algorithm based on improved CenterTrack is proposed,and an attention mechanism is added to the backbone network to improve the detection accuracy of the feature map.The tracking algorithm of enhanced detection is designed to enhance the detection effect of the next frame while completing the multi-target tracking,so as to improve the tracking accuracy.The generalization ability of the algorithm is improved by training the data set processed by the data enhancement method.Verified on MOT17 dataset,experimental results show that the improved algorithm has higher tracking accuracy and correct detection ratio than the original method.