基于全景的CMVision算法改进

Algorithmic Improvement for CMVision in Omnidirectional Image Processing

针对全景图像存在大量无效像素的特点,通过避免无效像素的处理使其处理效率得到大幅提升,同时也避免了一些颜色识别的干扰;针对全景图像中心对称的特点,通过改进描述目标的数据结构使其能够更加合理而准确地表示目标的位置信息,并通过查表的方法高效地实现从图像中提取这些信息,从而为后续的路径规划和避障算法奠定良好的基础。通过试验,对比改进前后的CMVision的性能表现,证明改进确实提高了类库在处理全景图像上的效率,获得了更准确的目标信息。

In light of the characteristic that the omnidirectional image has too many invalid pixels, this paper improves the performance of arithmetic prominently through avoiding processing these invalid pixels, at the same time, avoiding some confusions of color recognition. In view of the characteristic that the omnidirectional image is centrosymmetric, the paper improves the information data structure of regions to make it describe the position of objects more reasonably and exactly, and extract these information from the image efficiently through table - search technology, which paves the way for the robot＇s path planning and obstacle avoidance. The experiment based on CMVision library proves that the improved library executes more efficiently and obtains more exact object information compared with old version in omnidirectional image processing.