A SIFT-Based Method for Matching Desired Keypoints on Mars Rock Targets

One of the goals of planetary exploration is to cache rock samples for subsequent return to Earth in the future Mars Sample Return (MSR) mission. The paper presents a new approach to determining matching points of any desired target points on a rock using a SIFT-based algorithm in the pair of images. Finding corresponding matching points in the pair of images is an extremely valuable capability for calculating the size and height of a rock in an autonomous planetary science exploration mission. This new method consists of four major processing procedures: Rock Detection, SIFT Processing, Triangulation, and Correlation. The first processing block rock detection has been designed to find a rock’s closed contour using the Canny algorithm and multiple levels of a multi-scale image pyramid in the pair of images. SIFT is used to compute matching points within the closed rock contour. In triangulation, we utilize known matching points found by SIFT to remove image rotation between the left image and the right image, then using the angle of the desired interest point and a Euclidean distance from the desired point to the nearest matching point found through SIFT in the left image to affirm a rough matching point in the right image. Correlation is exploited to make sure an accurate matching point is achieved with respect to the desired interest point in the right image. In our implementation a 7 x 7 correlation window is used with a threshold of 0.8 on the correlation score. After we have found desired interest points, we can work out the size and height of the rock target through known arguments of the camera parameters. Current work is in preparation for the eventual grasping of a rock using our developed technique. We are currently testing our program code (OpenCV) using image data acquired from the Aberystwyth University PanCam Emulator (AUPE) instrument of our Trans-National Planetary Analogue Terrain Laboratory (PATLab). Experimentation results are presented in this paper.