An algorithm for walking goal line detection was hereby proposed based on improved randomized Hough transform (RHT) for harvesting robot. A camera was fixed on the top of the combined harvester in order to capture images of farmland scenes in the process of harvesting. By virtue of statistic analysis and edge detection, the target location of walking goal line as well as the optimal points in the straight-line direction were determined according to different color features of harvested areas, non-harvested areas and areas without farmlands respectively. The optimal points selected as the point set, complete line detection was accomplished by applying RHT, which avoids invalid sampling and accumulation in comparison with the traditional algorithm of RHT. Through processing various pictures, it was well proved that the algorithm can effectively detects the linear parameters, with the processing time of about 200ms. 