Target Recognition Based on SIFT Algorithm for a Train Uncoupling Robot

The train uncoupling robot is a 4-DOF robot which can complete the uncoupling work efficiently and accurately in the process of disassembly or marshalling of freight trains or passenger trains without the help of workers. The uncoupling work should be completed when the train and the uncoupling robot are both in motion. However, this increases the difficulty of recognizing the hook of trains and finishing the uncoupling work automatically. Therefore, the visual system of the robot plays a vital role during the uncoupling process so that it can be able to recognize the target hook by analyzing images and data from cameras and sensors. Furthermore, it should also have the ability to predict the motion planning of the object. In this paper, a visual control scheme is developed by using the Scale-invariant Feature Transform (SIFT) algorithm to recognize the hook of train in motion. It searches the extreme points as the key-points, and extracts the features such as position, scale and direction to match between the two images. The results indicate that the target hook can be recognized accurately and the recognition speed can meet the working requirements of the system by using the proposed visual control scheme even if the size of the target is varying or the target rotates moderately.