Aimed at visual navigation of grain combine harvester, an algorithm was proposed for walking goal line detection on improved Hough transform (HT) on grain combine harvester. Through the improvement for image threshold segmentation of the 1-D maximum entropy, the computing speed was increased. After scanning the rows and columns of binary images, the end position of the walk goal line and the candidate points on the direction of the walk goal line were determined. Then the candidate points were used as points set, least squares methods and the end position of the goal line were combined to determine the point from the test line. The improved HT has finished the test line detection. Compared with the traditional HT, the improved HT transfers binary mapping into 1-D mapping, the computing speed was accelerated, and the space occupied was reduced. Meanwhile, the anti-interference ability was improved effectively. After testing thousands of pictures, the algorithm was well proved with effectively detecting the linear parameters, and the processing time was about 100ms.