一种融合深度信息的视频目标压缩跟踪算法

Video Target Compressive Tracking Algorithm with Fusion Depth Information

针对复杂环境下单一特征在跟踪过程中易造成准确率下降和鲁棒性差的问题,提出一种融合深度信息的视频目标压缩跟踪算法。利用压缩感知理论分别提取目标灰度图像和对应深度图像的正负样本压缩特征,通过特征训练弱分类器,利用马氏距离赋予弱分类器权值,加权组合为强分类器,实现目标的多特征融合,视目标跟踪为一个二分类问题,确定目标跟踪结果。使用由粗到细的搜索策略减小计算复杂度。实验结果表明,该算法跟踪目标平均中心位置误差为9.95像素,平均成功帧率可达96%,算法保持实时性的同时对视频目标运动遭遇的部分遮挡、姿态变化、光照变化以及相似物干扰等情况下的跟踪均具有较好的效果。

Aiming at the problem that single feature in the complex environment is prone to the decrease of the accuracy and the robustness in the tracking process,a video target compressive tracking algorithm was proposed to integrate the depth information.Compressed sensing theory was used to extract the positive and negative compressive features of the grayscale image and the corresponding depth image respectively,weak classifiers were trained by the feature,weights of the weak classifiers were given by the Mahalanobis distance,and combined to strong classifier to realize the multi-feature fusion,the target tracking was regarded as a binary problem to determine the target tracking result.A coarse-to-fine strategy is adopted to reduce the computational complexity.Experimental results show that the algorithm has an average center position error of 9.95 pixels,the average success rates up to 96%,the algorithm keeps real-time feature and performs well against occlusion,pose variation,illumination and background clutter.