基于背景约束的抗遮挡目标跟踪算法研究

Research on anti⁃occlusion target tracking algorithm based on background constraint

为获取高质量的目标跟踪结果,基于背景约束机制设计了一种新的抗遮挡目标跟踪算法。根据视觉纹理特征提取初始帧图像内的前景目标与背景特征,并通过建立前景目标物的概率约束提高目标识别的精度;采用粒子滤波跟踪框架,以颜色和二值形状为跟踪线索,利用Bhattacharyya距离判定目标物是否被遮挡;在目标遮挡条件下,采用自适应融合策略适时转变线索融合模式,同时将mean-shift算法引入粒子滤波框架内,实现抗遮挡目标跟踪。实验结果表明,非遮挡情况下的目标跟踪耗时在14 ns以内,遮挡情况下的目标跟踪耗时在17 ns以内;最大中心位置误差仅为16.7像素,最大覆盖率可达到82.1%,说明该算法具有较好的跟踪效果。

In order to obtain high quality target tracking results,a new anti⁃occlusion target tracking algorithm is designed based on background constraint process.The foreground object and background feature in the initial frame image are extracted according to the visual texture feature,and the target recognition accuracy is improved by establishing the probability constraint of the foreground object.The particle filter tracking framework was adopted.The color and binary shape were used as tracking clues,and the Bhattacharyya distance was used to determine whether the target object was blocked.Under the condition of occlusion,adaptive fusion strategy is adopted to timely transform the cue fusion mode,and mean⁃shift algorithm is introduced into the particle filter framework to realize the anti⁃occlusion target tracking.The experimental results show that the target tracking time in the case of non⁃occlusion is within 14 ns,and the target tracking time in the case of occlusion is within 17 ns.The maximum center position error is only 16.7 pixels,and the maximum coverage rate can reach 82.1%,indicating that the algorithm has a good tracking effect.