一种基于双目视觉传感器的遮挡目标期望抓取点识别与定位方法  

A Method for Recognizing and Locating the Expected Grasping Position of Occluded Target Based on Binocular Vision Sensor

作　　者：王煜升 张波涛[1] 吴秋轩[1] 吕强[1] WANG Yusheng;ZHANG Botao;WU Qiuxuan;LÜQiang(School of Automation,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)

机构地区：[1]杭州电子科技大学自动化学院,浙江杭州310018

出　　处：《传感技术学报》2021年第8期1033-1043,共11页Chinese Journal of Sensors and Actuators

基　　金：浙江省重点研发项目(2019C04018);国家自然科学基金(62073108)。

摘　　要：在不确定性较高的室内复杂场景中,机器人常需识别遮挡物体并对其抓取。遮挡问题会导致抓取点预估位置脱离目标,产生位置漂移。针对该问题,本文提出一种基于双目视觉的遮挡目标抓取点识别与定位策略。采用基于期望位置模型的方法估计,以特征检测进行遮挡目标识别,并进行轮廓还原;根据期望抓取点模型,采集目标的期望抓取位置,构建位置模型库。将待检测目标与模型库匹配,提取双目视野中未遮挡区域的期望抓取点。实验表明本方法在复杂环境下具有较高的鲁棒性,抗干扰能力强,对遮挡目标具有较高的定位精度。In the indoor complex scene with high uncertainty, the robot often needs to recognize and grasp an occluded object. The occlusion problem will cause the estimated grasping position to deviate from the target, resulting in position drift. To deal with this problem, A method for recognizing and locating the expected grasping position of occluded target based on binocular vision is proposed. A desired position model is designed to estimate the object, recognizes occluded object with feature detection and restores the contour. The expected model is used to estimate the position, and the feature detection is used to recognize the occluded object, and then the contour is restored;According to the above model, the expected grasping position is collected, and the position model database is constructed. The object to be detected is matched with the database to extract the desired grab points. Experiment results show that the proposed method has high robustness, strong anti-interference ability and high positioning accuracy for occluded object in complex environments.

关 键 词：机器视觉 抓取位置估计 遮挡目标识别 双目定位 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]