基于YOLO的目标人员身份识别与跟踪方法

Human Identification and Tracking Method Based on YOLO

在多人存在的单目标人员视觉跟踪场景中,目标人员的遮挡及丢失是常见的问题之一。针对目标由于遮挡等原因丢失后的恢复问题,提出了基于单目标检测与身份识别网络模型相结合的视觉跟踪方法。该方法基于YOLO模型进行候选目标检测,基于在身份识别数据库上预训练的残差网络,进行候选目标基础特征提取。通过计算候选目标特征与预提取的模板目标特征的均方误差和,进行前后帧的数据关联,达到跟踪目标的效果。实验表明,论文提出的方法在各种目标人员消失的场景下具有有效性。

The occlusion and loss of targets is one of the common problems in dynamic visual tracking scenarios with multiple false targets. Aiming at the recovery after the target is lost due to occlusion and other reasons,a visual tracking method based on the combination of single target detection and identity recognition is considered. This method is based on the YOLO model for target detection. The unique features based on residual network pre-trained on a human identification dataset are extracted. By calculating the sum of the mean square error of the between candidates and the target template,the target is identified and tracked between frames. Experiment results show that the method is effective in various cases of targets disappear.