基于SORT算法的图像轨迹跟踪混合控制方法

Image trajectory tracking hybrid control method based on SORT algorithm

当目标物体被其他物体部分或完全遮挡时,目标的有效特征点数量会逐渐减少,跟踪器无法继续准确地锁定目标,导致目标轨迹中断。为此,文中研究基于SORT算法的图像轨迹跟踪混合控制方法。选取FCOS算法,利用特征金字塔结构,依据检测头层输出的目标分类得分、位置回归结果以及中心度检测图像目标。将目标检测结果作为卡尔曼滤波器的输入,利用离散控制过程系统描述视频图像中的目标运动状态,预测目标轨迹。利用SORT算法控制图像目标检测结果与目标轨迹预测结果进行级联匹配与IoU匹配,输出匹配成功的目标,即图像目标轨迹跟踪结果。实验结果表明,该方法可有效地跟踪视频图像目标轨迹,未出现ID变更情况,轨迹中断占比低于0.2%。

When the object is partly or completely occluded by other objects,the number of effective feature points of the object will decrease gradually,and the tracker will not be able to target the object accurately,resulting in the interruption of the object trajectory.Therefore,an image trajectory tracking hybrid control method based on SORT(simple online and realtime tracking)algorithm is studied.The FCOS(fully convolutional one-stage object detection)algorithm is used to detect the image object based on feature pyramid structure according to the object classification score,the result of position regression and the center degree output by the layer of the detector.The object detection result is taken as the input of the Kalman filter.The discrete control process system is used to describe the object motion state in the video image and predict the object trajectory.The SORT algorithm is used to control the cascade matching and IoU matching between the image object detection result and the object trajectory prediction result,and output the successfully matched targets,that is,the image object trajectory tracking results.The experimental results show that the method can track the image object trajectory effectively without ID change,and its percentage of track interruption is below 0.2%.