基于多源信息融合的采摘机器人果实感知定位机制研究

项目名称： 基于多源信息融合的采摘机器人果实感知定位机制研究

项目编号： No.31301235

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 农业科学

项目作者： 卢军

作者单位： 华中农业大学

项目金额： 21万元

中文摘要： 机器人采摘是农业机械化和自动化的趋势，利用机器视觉进行检测定位是普遍采用的手段。由于水果的生长环境以及个体的形状和颜色不同,自然光照强度变化和阴影、遮挡等都会对图像处理和理解带来困难, 影响检测定位的准确率和采摘的成功率。基于树上水果检测定位时的复杂性，本研究以树上柑橘为研究对象，通过融合彩色图像和深度图像的方法，利用基于TOF技术的3D相机获取场景的深度图，通过深度信息来弥补传统的彩色图像检测目标时光照变化所带来的高光、阴影等一系列的问题，从而完整的分割出场景中的水果目标区域。并通过整合格式塔法则和水果形状先验知识的轮廓编组技术，完整的恢复出树上水果可能被遮挡的轮廓。通过基于TOF技术的3D成像技术和轮廓编组算法，增强对当前水果采摘机器人感知定位时所遇到的光照变化、阴影和遮挡等瓶颈问题的处理能力和适应能力，为提高采摘机器人感知定位的鲁棒性和自适应性提供新的解决方案。

中文关键词： 图像融合；轮廓整合；遮挡恢复；树上水果检测；

英文摘要： Fruit harvesting with robots is the trend of agricultural mechanization and automation. How to locate and characterize the fruits on the tree is a key problem,it's a widely-used method to detect and locate fruits with machine vision. However, occlusion, variable illumination, diverse growth enviroment, variable appearance and texture make this task a complex challenge. This research focuses on detection and localization of the citrus fruits from the cluttering background. Because of the complexity and uncertainty of fruits detection, the image fusion method between depth map and color image was used, the range map which is acquired by a 3D camera based on TOF(Time of Flight) technology was used to make up these problems such as highlight, shadow, variable illumination, so as to detect the object region of fruits completely and accurately. The contour grouping method, it's intedgrated of Gestalt laws and prior knowledge of fruit shape, was used to recover the whole contour which was usually occluded by others. A new solution by image fusion method, 3D imaging technology and contour grouping algorithm was proposed to improve the robust and adaptive ability for detection and location of fruits within the tree.

英文关键词： Image fusion；Contour integration；Occlusion recovery；Detecting fruit on tree；