Study On Calculation Method Of Number And Height Of Trees Based On Computer Vision

"Clear water and green mountains are gold and silver mountains",forestry has important research value and significance in environmental protection.With the advent of the computer age,computer vision technology has been widely used in various industries,if the application of computer vision in forestry engineering,forestry development has an important breakthrough significance.Computer vision is changing the traditional forestry research methods and forestry information engineering technology is becoming mature.In view of the problems of low accuracy,difficulty in operation and difficulty in transforming professional knowledge into rules in forestry engineering,it will be a better choice to use the theory,mechanism and method of computer vision to replace manual operation and realize non-contact measurement calculation.In order to better promote the progress of forestry information.In this paper,traditional binocular vision algorithm and deep learning target detection algorithm in computer vision are used to innovate the fusion of traditional vision and modern vision,and a method of tree number statistics and tree height measurement based on computer vision is designed and implemented.Due to the wide variety of trees,this paper mainly selects typical trees with coniferous and broad leaves as experimental data.This paper is mainly studied in the following aspects:(1)Aiming at the problem of calculating the number of trees in the video screen,a statistical method of the number of trees based on target detection is proposed.Firstly,data acquisition,including image data and video data,is carried out.The shooting method is divided into overlooking shooting and side shooting.Labelme is used to filter data and make labels.YOLOv4 target detection framework was selected to detect trees in the picture,and the trained weight was obtained.In the video screen,"line collision method" was used to calculate the number of trees.(2)Aiming at the problem of non-contact measurement of tree height,this paper proposes a tree height measurement method based on binocular vision.Using binocular camera as data acquisition equipment,the measurement of tree height requires knowing the spatial coordinates of the contact points between the top and root of trees and the ground,and the shooting method is side shooting.Data were taken from representative trees,including coniferous and broadleaved,single and multiple trees,with and without leaves.In view of the existing problems such as tree shelter,research.BM algorithm and SGBM algorithm are used for parallax matching of trees in binocular image,and the spatial coordinates of key points of trees are obtained to calculate the height of trees.(3)For the problem of matching the "specified" target tree in binocular image,but not the whole image.A tree height measurement method based on binocular vision and deep learning was proposed.This part is divided into innovative points of this paper.Binocular vision belongs to traditional vision category,while deep learning belongs to modern vision category.A tree height measurement method combining binocular vision and target detection is designed to detect the target of trees in binocular image,and parallax matching is carried out for target trees in the "detection frame",which has high research value.