Kinect传感器的植株冠层三维数据测量

摘要

植株三维信息重构能为植株生长状态监测和精确喷雾施药提供有效数据。提出一种基于Kinect传感器技术的植株冠层三维数据测量的方法。由Kinect传感器进行植株彩色和深度图像数据的采集，提取和处理所采集的植株冠层目标有效三维信息，完成对植株深度数值和水平投影面积的计算。以规则形状物体与不规则植株为实验对象，对三维数据测量方法进行准确性实验测试，并将实验结果与人工测量结果进行比对。实验结果显示，该方法的深度和面积测量的准确性较高，深度测量误差小于1.0%，面积测量误差小于3.6%。选取温室吊兰作为场地实验对象，采用由测量机构和控制处理机构组成的冠层三维检测系统对吊兰冠层进行三维数据测量，并实时输出深度以及水平投影面积信息，其深度测量的相对误差为1.77%。研究表明，该方法具有较高的可行性，适用于温室植株冠层三维数据测量。%The image reconstruction for three dimensional (3D) plant structures could be used to monitor plant growth automatically and provide real-time spray amount information for precision agriculture. A 3D image reconstruction method based on Kinect sensor technology was proposed to measure canopy stereoscopic structures. The color images and depth images of plants were detected with a Kinect sensor at the same time. The depth distances between objects and sensor were obtained based on average calculations after multiple depth image measurements. The horizontal projection area of plants and 3D canopy structure reconstructions of plants were achieved by extracting useful data and fusing color image and depth image information. Some regular-shaped objects and irregular-shaped plants were chosen as scanning targets to test the accuracy of the proposed method for depth distance and horizontal projection area estimation. The experiment results showed that the new method could detect both regular-shaped objects and irregular-shaped plants accurately with depth distance error less than 1.0% and horizontal projection area error less than 3.6% in all three depth detection distances from 1.1 m to 1.3 m. A planting area for potted chlorophytes in a greenhouse were used as scanning targets to verify the performance of the new method for detecting depth distances, measuring horizontal projection area, and reconstructing 3D canopy structures in real time. A 3D image detection and reconstruction system integrated with a Kinect sensor measuring mechanism and a position controller was used to conduct the experiments. The experiment results showed the average depth distance error was less than 1.77% and the proposed method has a high feasibility for 3D canopy structure measurements of greenhouse plants.