基于深度卷积CNN的采摘机器人动态避障方法研究  

Research on Dynamic Obstacle Avoidance Method of Picking Robot Based on Deep Convolution CNN

作　　者：张茉莉 Zhang Moli(Quanzhou University of Information Engineering,Quanzhou,Fujian 362011,China)

机构地区：[1]泉州信息工程学院,福建泉州362011

出　　处：《黑龙江工业学院学报（综合版）》2024年第7期122-129,共8页Journal of Heilongjiang University of Technology(Comprehensive Edition)

基　　金：福建省中青年教育科研项目(科技类)“基于无意识设计理念的急救装备设计研究”(项目编号:JAT190934)。

摘　　要：种植园环境除了静态障碍物外,还存在较多动态障碍物,干扰采摘机器人行动。为了提升采摘机器人动态避障能力,引入深度卷积CNN,提出一种新的避障方法。以立体视觉系统实时感知外界环境,获取深度图像;利用深度卷积CNN提取深度图像特征,识别动态障碍物,根据机器人与障碍物之间的相对速度进行适当调整,以保持安全距离,控制机器人移出碰撞区域,实现动态避障。实验结果表明:所研究的采摘机器人和障碍物未发生碰撞行为,且每一次相遇时采摘机器人与动态障碍物之间的实际距离均大于1m安全距离限值。由此证明,研究方法具有有效性。In addition to static obstacles,there are also many dynamic obstacles in the plantation environment,which interfere with the action of the picking robot.In order to improve the dynamic obstacle avoidance ability of picking robot,a new obstacle avoidance method is proposed by introducing deep convolution CNN.Real-time perception of external environment with stereo vision system to obtain depth image;use depth convolution CNN to extract depth image features and identify dynamic obstacles;according to the relative speed between the robot and the obstacle,the robot is properly adjusted to keep a safe distance,and the robot is controlled to move out of the collision area to realize the dynamic obstacle avoidance.The experimental results show that there is no collision between the picking robot and the obstacle under the research method,and the actual distance between the picking robot and the dynamic obstacle is greater than the 1m safety distance limit at each encounter.This proves that the research method is effective.

关 键 词：深度卷积CNN 采摘机器人 立体视觉系统 动态避障方法 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]