Crop-row Recognition Based On Line Detection

The autonomous navigation of agricultural machinery is very important for the development of modern agriculture.Since crops in the field are arranged in line,crop rows include the guiding information.The research about methods to detect crop row is crucial which can provide the guiding parameters for agricultural machinery,so that the farming efficiency is improved.The aim of this study is to present a general method for multiple row detection.In this paper,image preprocessing,detection and classification of crop rows center,line detection are studied using the technology of line detection and the theory of digital image processing.Firstly,the original color image is segmented by excess green index and iterative threshold and then filtered by continuous expansion and corrosion filtering operation,so that not only the information contained in the image is reduced,but also can we extract the areas of crop rows with less influence of noise.Secondly,points indicating center of the rows of each strip was detected and classified by the improved strip analysis based on based on the regularities of first-row's distribution and matching model.So that a general method for multiple row detection is presented which solve the problem of traditional vertical projection method s can detect only one row in the center.Finally,the estimation of the row center lines is accomplished by least squares regression analysis and probabilistic Hough transform.Compared with standard Hough transform,probabilistic Hough transform could reduce the amount of computation and improve the efficiency of algorithm.Experiments under 50 various images have proved the proposed method to be a reliable crop row detection method which can detect multiple crop rows and also is suitable for the various conditions of rocks,debris and shaded area in the farmland.In addition,points indicating center of the rows were accurately detected and classified,also the crop rows detected reflect the location with the advantage high level of feasibility and short computational time.To process a 400 ? 247 pixel image,the average time is 178.68 ms which could meet the practical requirements of farmland.