自然光线环境中的空间物体快速识别和定位算法研究

针对智能机械臂在自然光环境的三维空间中对目标物体的自主识别率和定位精度低的问题,提出了一种基于深度学习的视觉和光学雷达融合定位算法,实现自然光线下空间物体的高精度快速定位。首先,采集 RGB 图像和深度数据,利用深度学习算法对图像进行目标识别与实例分割;然后,将实例分割目标物的二维深度矩阵转换成三维空间点云;最后,用综合修正算法对位置修正,实现对目标物体在三维空间的抓取位置精准定位。 通过不同光照强度下的目标物体识别和定位实验验证了该算法的有效性和实用性,获取的目标物体的三维空间坐标较为精确,单位距离的定位误差在 0. 5%以内,受照明亮度影响较小,对机械臂智能抓取的研究具有较为重要的意义。

Recognition and Positioning Algorithm of Space Objects in Natural Light Environment

Aiming at the problem of low autonomous recognition rate and low positioning accuracy of target objects in the three-dimensional space of natural light environment,a deep learning-based vision and optical radar fusion positioning algorithm is proposed to achieve high precision of space objects and rapid positioning under natural light. Firstly,deep learning algorithms is adopted to perform target recognition and instance segmentation of the image by collecting RGB images and depth data. Then,the two-dimensional depth matrix of the instance segmentation target is converted into a three-dimensional point cloud. Finally,a comprehensive correction algorithm is applied to correct the position for achieving precise positioning of grasping position of the target object in three-dimensional space. The effectiveness and practicability of the proposed algorithm is verified through target object recognition and positioning experiments under different light intensities. The three dimensional space coordinates of the target object obtained are more accurate,and the positioning error per unit distance is within 0. 5%,which is less affected by illumination brightness. The research on intelligent grasping of robotic arms i of great significance.