A YOLACT++-Based System for Automated Tree Height Estimation

Tree height is an important indicator for describing tree growth characteristics. It has a significant impact on calculating timber volume and evaluating site and forest quality. Over the past few years, the use of popular smartphones to measure standing tree factors has attracted the attention of some scholars. The existing mobile phone-based standing tree factor measurement methods do not realize the automated measurement of standing tree factor and require complicated operations, which does not take advantage of the convenience of mobile phones. With the development of computer vision, using computer vision to solve the problem of standing tree factor measurement has become a new idea. Therefore, we proposed a standing tree height measurement method based on YOLACT++ and created an automated measurement App. First, ARCore Depth API and Midas were used for depth estimation. Secondly, YOLACT++ was used for standing tree image segmentation. Subsequently, 3D reconstruction was carried out using RGB images and depth maps, followed by extracting the point cloud of the target tree based on the segmentation results. Finally, the predicted height of the tree can be obtained. After verification, the accuracy of standing height measurement reached 96.32%, which met the accuracy requirements of forest measurement.