基于视差图的未知环境下农田障碍物检测方法

摘要

Considering the irregularity of obstacles as well as the variability of background in intelligent agricultural robot working environment,a disparity-based method for obstacle detection in unknown environment is proposed.Two inexpensive web-cams are adopted to capture left and right images,based on which the initial disparity image is calculated.Subsequently,the remote scene is filtered by a disparity threshold,and then potential obstacles are obtained from the disparity image through region connection.The method is finalized by the recognition and location of obstacles via the filtration of area and height threshold.Experimental results indicate that the presented algorithm could bypass the prior information of obstacles and background; In the distance of 5m from the camera,the absolute error percent of the calculated distance is limited below 4.37 %,with that of the calculated size within 3.81%.%针对智能农业机器人作业环境中障碍物不规则和背景受自然光照等外界环境干扰大的问题,提出一种基于视差图的未知环境下障碍物检测方法.采用价格低廉的网络摄像头采集左右场景图并计算对应的视差图；用视差阂值过滤掉远处的景物,再通过区域连通,得到独立的潜在障碍物；最终经面积闽值和高度阈值的过滤,检测出障碍物并定位.实验结果表明,该检测算法不需要障碍物和背景的先验知识,在距离摄像头5m视野内,障碍物距离和尺寸测量的相对偏差绝对值分别控制在4.37％和3.81％以内.