摘要
提出了一种利用激光测距在球坐标系中快速检测障碍物边缘的新算法—归一化距离差自动阈值法.首先通过假设每一像点的坡度为零而求得此像点的期望归一代距离差,再对实际归一化距离差与此期望归一化距离差的商取阈值而检测障碍物边缘.本文还系统分析了激光测距成像雷达的测距周期性问题对边缘检测结果的影响,并提出了消除由此问题而引起的边缘检测误差的有效方法.
This paper presents a new method for rapidly detecting the edges of obstacle in spherical coordinate using range image sensed by Laser Imaging Range Sensor(LIRS). The central idea of this algorithm which automatically thresholds the normalized range differences is: the expected normalized range differences based on the slope being zero are first obtained, then edges of obstacle are detected by thresholding the quotients between actual normalized range differences from range image and these expected normalized range differences. The algorithm is shown to have rapid and robust results for edge detection of obstacle of real outdoor range image, and is very useful for obstacle detection in Autonomous Land Vehicle(ALV). In addition, the errors of edge detection caused by the periodicity problem of range sensing of LIRS is analyzed in detail, and a strategy known as ' offsetting ranges ' to eliminate these errors is also presented.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1998年第2期154-160,共7页
Pattern Recognition and Artificial Intelligence
关键词
距离图
边缘检测
算法
激光测距雷达
Range Image, Edge Detection of Obstacle, Slope, Normalized Range Differences, Periodicity Problems of Range Sensing
