一种基于概率神经网络多信息融合的移动目标跟踪算法

A Multi-cue Fused Moving Object Tracker Based on Probabilistic Neural Networks

采用概率神经网络(PNN)实现了对图像序列中移动目标——人头的跟踪.由于采用单一特征信息的跟踪算法在复杂环境中往往失效,故以头部的颜色信息模板和头部轮廓的梯度信息模板作为跟踪依据,并通过改变PNN的结构实现了图像信息的融合以及自适应模板修正.实验结果表明,基于PNN的算法在处理目标的旋转和遮挡时有着良好的效果,且具有简单、跟踪鲁棒性好等特点.

Probabilistic neural networks were used to track a moving person＇s head in image sequences. Color and contour of the head are chosen as prototypes because of the deficiency of tracking based on solo cue. In order to perform an image cue data fusion algorithm the PNN architecture is adapted accordingly. The algorithm tolerates rotation of the object as well as occlusion. Compared with other tracking methods, the PNN approach is simpler and more robust.