Abstract: To realize the space positioning as well as the precise detection to doors for robots,the binocular visual based door detection method was proposed.The vertical line segment in images captured by binocular cameras was extracted based on Scharr operator and Hough transform.Based on the similarity in feature histogram,a list of matched line segments pairs which represented the same line in the real world was detected.According to the principle of binocular measurement method,the concept of door candidate was proposed,and the features of door side frame,depth mutation,occlusion and color symmetry were extracted.The  multi-variate features of doors was effectively coped with the characteristic integrating system,and doors in indoor environment were detected.Experimental results showed that the proposed method could perform door detection with accuracy rate over 93%.Moreover,it was fast enough to be used for real-time applications in mobile robot systems.

Key words: vertical line segments matching, door detection, binocular vision, mobile robot systems

摘要： 为了提高家居环境下门识别的准确率和门定位的精度,提出一种基于双目视觉的门识别方法。该方法通过采用Scharr算子和霍夫变换算法提取双目传感器采集图像中的竖直线段,并利用特征直方图相似度进行竖线特征匹配。根据双目测距原理提出“疑似门”的概念,提取了门的侧边框特征、深度突变特征、遮挡特征和颜色对称特征,利用特征积分制有效地解决了门的状态多变问题,实现了对家居环境下门的识别。实验结果表明,该方法对门的识别率可达93%,同时检测速度符合家庭服务机器人对实时性的要求。

关键词: 竖线特征匹配, 门识别, 双目立体视觉, 家庭服务机器人