基于Kinect V3传感器的叶菜类作物三维重建与表型参数获取

Three-dimensional Reconstruction and Phenotypic Parameters Acquisition of Leafy Vegetables Based on Kinect V3 Sensor

摘要:

作物三维重建是实现作物表型量化和精准获取的有效手段，可为育种和栽培提供基础数据支撑。本文提出了一种基于Kinect V3传感器的叶菜类作物三维重建与表型参数无损获取方法。首先，设计了一种可实现作物多视角点云快速采集的低成本三维重建平台，其载物台面设计成多个标定点，可利用台面信息进行点云水平校准。其次，采用载物台恢复与广义迭代最近点（Generalized iterative closest point，GICP）算法相结合的方式对获取的多视角点云进行配准拼接，实现叶菜类作物三维重建。最后，借助有效的表型参数测量，实现对叶菜类作物株高、叶长、叶宽、叶面积等表型参数的精准获取。为评估该方法相似度，选取木耳菜、甘蓝、茄子、紫背天葵的苗期植株为试验对象，将其与SFM-MVS方法进行对比。试验结果表明，木耳菜、甘蓝、茄子、紫背天葵点云间平均距离误差分别为0.381、0.340、0.195、0.270 cm，二者的三维重建结果具有较高相似度。与人工实测值相比，借助该方法提取木耳菜和紫背天葵株高、叶长、叶宽、叶面积决定系数均不低于0.903，平均绝对百分比误差不高于9.759%，木耳菜和紫背天葵株高、叶长、叶宽、叶面积均方根误差分别为0.366 cm、0.203 cm、0.290 cm、3.182 cm2和0.496 cm、0.344 cm、0.282 cm、0.825 cm2，表明其具有较高测量精度。上述方法可为设施农业育种和栽培提供快捷、高效的作物表型获取途径。

Abstract:

Crop three-dimensional reconstruction is an effective means to realize crop phenotype quantification and accurate acquisition, and can provide basic data support for breeding and cultivation. A nondestructive acquisition method for three-dimensional reconstruction and phenotypic parameters of leafy vegetable crops were presented based on Kinect V3 sensor. Firstly, a low-cost three-dimensional reconstruction platform that can realize rapid acquisition of multi-view point clouds of crops was designed. The loading surface of the platform was designed as multiple calibration points, and the table surface information can be used for point cloud horizontal calibration. Secondly, the multi-view point clouds obtained were registered and spliced by combining the carrier platform restoration and the generalized iterative closest point (GICP) algorithm to realize the three-dimensional reconstruction of leafy vegetable crops. Finally, through effective phenotypic parameter measurement, the accurate acquisition of phenotypic parameters such as plant height, leaf length, leaf width, and leaf area of leafy vegetable crops was achieved. To evaluate the similarity of this method, seedling plants of Malabar spinach, cabbage, eggplant, and purple back sunflower were selected as test objects and compared with the SFM-MVS method. The test results showed that the average distance errors between the point clouds of Malabar spinach, cabbage, eggplant, and purple back sunflower were 0.381 cm, 0.340 cm, 0.195 cm, and 0.270 cm respectively, and the three-dimensional reconstruction results of the two had high similarity. Compared with the manual measured values, the determination coefficients of plant height, leaf length, leaf width, and leaf area of Malabar spinach and purple back sunflower extracted by this method were not less than 0.903, and the average absolute percentage error was not higher than 9.759%. The root mean square errors of plant height, leaf length, leaf width, and leaf area of Malabar spinach and purple back sunflower were 0.366 cm, 0.203 cm, 0.290 cm, 3.182 cm2 and 0.496 cm, 0.344 cm, 0.282 cm, 0.825 cm2, respectively, indicating that it had high measurement accuracy. The above method can provide a fast and efficient way for crop phenotype acquisition for facility agriculture breeding and cultivation.