激光视觉融合下的运动检测与失配矫正

发布时间：2018-04-08 22:04

本文选题：激光与视觉融合　切入点：运动检测　出处：《光电工程》2017年11期

【摘要】：针对单一传感器在动态场景感知问题上的局限性,设计了一种融合激光与视觉的实现系统,并对运动检测中的背景显露区误判问题和融合中不同传感器间点云的失配问题分别提出了改进算法。在运动检测上,首先基于视觉的背景差分算法对激光进行前景点分拣,再以激光前景点为启发信息进行视觉前景聚类。在融合失配问题上,首先基于栅格失配度分别对激光和视觉点云进行聚类分割,再以激光为基准,逐一将对应的视觉点云与之配准,滤除噪声后所得到的矫正点云可用于场景重建进行进一步验证。实验结果表明,改进算法所获得的融合前景对"影子"有更好的鲁棒性;较之整体配准的矫正,改进算法在平均失配度上降低了约75%,在y和z方向上的偏移比收敛了至少5%。
[Abstract]:Aiming at the limitation of single sensor in the problem of dynamic scene perception, a realization system combining laser and vision is designed.An improved algorithm is proposed for the mismatch of the background exposed region in motion detection and the point cloud mismatch between different sensors in fusion.In motion detection, the background differential algorithm based on vision is first used to sort out the spot before the laser, and then the vision foreground clustering is carried out with the laser foreground as the heuristic information.In the fusion mismatch problem, the laser and the visual point cloud are segmented based on the grid mismatch, and then the corresponding visual point cloud is registered one by one based on the laser.The corrected point cloud after filtering noise can be used to further verify the scene reconstruction.The experimental results show that the fusion foreground obtained by the improved algorithm is more robust to "shadow", and the average mismatch degree is reduced by about 75.5%, and the migration ratio in y and z directions converges at least 5% compared with the correction of global registration.
【作者单位】：中国科学技术大学自动化系;中国科学院合肥物质科学研究院先进制造技术研究所;
【基金】：国家科技支撑项目(2015BAK06B02) 江苏省科技支撑项目(BE2013003) 国家自然科学基金(61401437)
【分类号】：TP391.41

【相似文献】