基于多特征的实时立体视觉检测方法

摘要

To make industrial robots and intelligent mobile platform available in the field of high-tech field which needs precise assemblage and operation,a high-precision stereo vision detection approach was proposed.Laser features were used in Harris corner detection,therefore feature detection was much more precise.The rude level detection was done by image pyramids struc-ture features based on direction vectors of images which got from two low resolution cameras together with line laser features from a high-speed laser scanning equipment.High-precision detection was achieved using both the displacement information of the laser and the image ROI (region of interest)features based on priori knowledge got from a high resolution monocular camera. Compared with the traditional method,the proposed method generates more accurate feature extraction result with less time con-sumption of the detection algorithm.%为实现传统工业机器人、智能移动平台的精密装配和精细操作，提出一种高精度的实时立体视觉检测方法。在低分辨率的双目立体视觉系统进行 Harris 角点特征提取时，融合激光传感器特征，提高角点检测的精确度，使用基于图像轮廓方向向量的图像金字塔特征结合激光线成像特征的方法，进行目标粗略检测，通过激光位移传感器提供位移信息，结合元件感兴趣区域的先验知识特征控制高分辨率的单目相机进行精确检测。实验结果表明，与其它算法相比，该方法具有较高的计算效率和准确性。