A Positioning Method for Apple Fruits Based on Image Processing and Information Fusion. Issue 17 (2018)

• Record Type: Journal Article
• Authors: Zheng, Lihua; Ji, Ronghua; Liao, Wei; Li, Minzan
• Is Part Of: IFAC-PapersOnLine. Volume 51:Issue 17(2018)
• Page Start: 764
• Page End: 769
• Publication Date: 2018
• Subjects: Apple orchard -- Three-dimensional camera -- GPS -- Attitude sensor -- Positioning; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2018.08.103
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: It is the key technology to position orchard fruits accurately and fast for yield forecasting and fruit harvesting. Aiming at forecasting apple orchard yield based on trees' images, a positioning method for apple fruits was proposed based on image processing and information fusion. Firstly, a position information acquisition system of orchard fruits was designed and established, which consisted of a three-dimensional camera, a GPS (Global Positioning System) receiver and an attitude sensor, and a world coordinates calculation method for apple fruits was developed according to this system. In the information acquisition system, the three-dimensional camera was used to acquire both the RGB image and depth image of apple trees in apple orchard; the GPS receiver was used to get the position information of the three-dimensional camera; the attitude sensor was used to obtain the attitude information of the three-dimensional camera. Secondly, a relative positioning calculation model was deduced by fusing the RGB images and the corresponding depth images, which was used to calculate the coordinates of the centers of each apple in camera coordinate system. Finally, a three-dimension coordinate converting method was investigated, and the GPS and attitude information of the camera was added to establish the absolute positioning calculation model to calculate the world coordinates of apples in the world coordinate system. Thus, each apple got only one space position in the world coordinate system, which could help to avoid counting the same apple in different images repeatedly. The experimental results showed that the mean relative error of the world coordinate positioning was 0.0375m; for the absolute error of the world coordinate positioning, the mean distance error of the longitude was 0.117m, that of the latitude was 0.437m, and that of the altitude was 0.145m. It indicated that the proposed method in this paper could be used to calculate the world coordinate for each apple in the orchard and its accuracy mainly depended on the detectors' accuracies.