结合分段复合权值与多策略的视觉运动目标跟踪

Visual tracking of moving objects based on piecewise fusion weight and multi-strategy

由于视觉监控中运动目标跟踪的准确性易受遮挡、摄像机运动、目标外观变化等因素的影响,本文提出了一种结合分段复合权值与多策略的视觉跟踪算法。该算法首先利用目标、背景以及候选区域特征信息建立分段的复合权值得到目标的位置概率分布。然后结合空间一致性和滞后阈值分割目标位置概率图以进一步抑制噪声干扰,同时通过分析分段复合权值变化判断目标遮挡,调整目标跟踪候选范围,并结合目标历史尺度信息对当前目标尺度进行自适应调整。最后,对目标以及背景区域信息进行动态更新以适应目标外观与场景变化。与典型算法进行的对比实验结果表明:该算法能够有效地应对目标遮挡与摄像机运动等因素的影响,实验时对各组视频的平均处理时间约为10ms左右,适用于复杂场景条件下运动目标的实时跟踪。

As the accuracy of moving object tracking in video surveillance is disturbed by occlusion,camera moving and target appearance changing,an algorithm based on piecewise fusion weight and multi-strategy was proposed.Firstly,the piecewise fusion weight was constructed by combining the feature of object,background and candidate regions to obtain the likelihood image of object location.Then,the likelihood image was segmented with the spatial coherent and hysteresis threshold to suppress noise interference.Meanwhile,the object occlusion was determined and handled by analyzing the change of the piecewise fusion weight and enlarging the candidate area.Furthermore,the object scale was adaptively adjusted according to history and current scales.Finally,object information and background regions were dynamically updated to adapt to the object appearance and scene changing.Experimental results compared with other traditional methods show that the proposed algorithm is applicable to process the moving object tracking in low-contrast scenes in real time,and the average processing time for different video images is 10 ms,which means that the algorithm is suitable for themoving object tracking in complex scenes.