视觉定位是移动机器人定位的一个主要发展方向, 针对传统视觉定位技术实时性差的问题, 提出一种基于自适应下采样的快速视觉定位技术。通过预先得到的尺寸特征, 并根据最小分辨尺寸计算下采样率, 而后对下一副图像进行下采样及图像分割, 根据对象坐标和下采样率确定对象在源图像中所处区域, 对源图像该区域进行图像分割和特征提取。将提取的尺寸特征作为下一幅图像的输入, 提取的视觉定位所需特征用于机器人的定位解算, 由此在保证视觉定位精度的前提下有效减少视觉定位的时间。实验表明, 对文中给定图像, 传统方法处理100幅图像时间为20.23 s, 而文中所述技术对应图像处理时间为1.78 s, 仅为传统技术的8.8%, 有效减少了图像处理时间, 提高了机器人视觉定位的实时性。

Visual locating technology is one of main development direction of mobile robot locating. Aiming at problem of poor real-time performance of traditional visual positioning technology, this paper proposes a fast visual locating technology based on adaptive down sampling. Down sampling rate is calculated firstly based on previous size characteristic and minimum resolution size. Secondly down sampling and image segmentation are made on next image. Object locating zone in original source image are determined according to coordinates of object and down sampling rate. Then image segmentation and feature extraction of source image of the area are made. Size of feature extraction is taken as an image input, and required features extraction of visual location are taken as robot locating solution. Thus visual positioning time is effectively reduced with ensured visual positioning accuracy. Experiments show that it takes 20.23 s for traditional technology to process 100 figures while it takes 1.78 s for technology given out by this paper, which is only 8.8% of traditional technology. This technology can reduce image processing time effectively and improve real-time performance of mobile robot locating.