摘要
针对传统的基于检测的在线目标跟踪算法容易产生跟踪漂移的现象,提出了一种新的在线目标跟踪算法。以基于主方向模板特征的双级联随机森林分类器作为检测器,卡尔曼滤波器作为跟踪器。首先利用卡拉曼算法跟踪目标,然后以跟踪的目标位置为中心向外扩展一定的范围作为双级联随机森林分类器的检测区域,利用全局随机森林分类器和局部随机森林分类器进行目标检测,并将检测结果作为Kalman跟踪算法下一帧的观测值。实验结果显示,提出的算法在跟踪大小420×320的图像时,跟踪速度达到24.3 f/s(帧/秒),目标中心位置误差在30 pixel时,算法准确率可达到80%以上。
As the traditional online tracking algorithm based on detection is easy to cause the tracking drift, a new online target tracking algorithm is proposed in this paper, where the Cascaded Random Forest with Dominant Orientation Templates is used as a detector, while the Kalman filter is the tracker. First, the Kalman filter is used to track the target, then the holistic detector and patch-based detector is applied to detect the object with the track result as the area center, and the detecting result is used as next frame＇ s observed value of Kalman tracking algorithm. The experimental results show that in the video sequence of 320 pixel × 240 pixel, the speed can keep in 24.3 frame/s, and the object center position error is in 30 pixel, while the accuracy can reach above 80%.
出处
《电视技术》
北大核心
2016年第12期23-27,共5页
Video Engineering
基金
国家自然科学基金项目(61301233)
