Horticultural Crop Point Clouds (Version 2)

Robust, multipurpose 3D plant model datasets are critical for the development of phenotyping and structural characteristic estimation methods. The study for which this dataset was produced details the construction of a multi-crop 3D point cloud dataset collected in-situ, tests the efficacy of HSV thresholding on colorized point cloud data to segment target plants, and evaluates the efficacy of mobile device LiDAR sensors for plant structural data collection. Across the 2023 agricultural season, mobile device LiDAR and RGB sensors were used to collect 295 colorized point clouds of pepper, tomato, and watermelon plants using the 3D scanning software Scaniverse. The vegetative plant tissue in each point cloud was segmented from its surroundings using HSV spectral thresholding and statistical outlier removal. Processing efficacy was evaluated through pointwise annotation of the raw point clouds and comparison with processed point clouds. Spatial accuracy of the mobile device sensors were evaluated via comparison of physically measured plant height and sensed plant height. Evaluation of the sensors' efficacy to measure plant structural characteristics was conducted by analyzing the error in fitted regressions between point cloud volume and physically measured structural characteristics. All processing and analysis code may be found in the following repository: hort-lidar-processing. The dataset is notable in that all plants were scanned in-situ, cultivated according to regional commercial production practices. An immediately apparent potential use for this dataset would be in the development and training of point cloud segmentation and phenotyping algorithms that are intended to function outside of controlled environments, using the tools and technologies available to growers.