摘要： 棒材出厂前要按根数打捆，基于单视角视觉识别计数方法难以处理多个目标之间的互遮挡等问题。文章提出多视角信息融合方法，从2个视角分别捕获目标图像，分别识别，再进行融合，对单视角目标识别信息的补偿与纠错，达到消除互遮挡、准确计数的目的。实践证明，该方法能很好地解决单视角的遮挡目标识别问题，对于因存在遮挡和光照条件限制而难以处理的多目标识别和跟踪问题具有一定的普适性。

关键词: 目标识别, 多视角, 信息融合, 遮挡目标, 棒材计数

Abstract: Steel bars need to be packed according to a standard amount. It is difficult for the counting method of single camera object recognizing to handle problems like target-hiding. This paper proposes a new method of multi-camera data fusion to solve this problem, capturing target images from two cameras and recognizing them respectively, then fusing the recognizing results. This is to compensate and correct the single camera recognizing result and meet the goal of eliminating hiding problem. Experiments show that this method can solve target-hiding problem well, and can be applied to multi-target recognizing and tracking problem caused by occlusion and light limitation.

Key words: object recognizing, multi-camera, information fusion, hiding targets, steel bars counting

中图分类号: 

• TP391