The NASA-ESA Mars Sample Return campaign would require transferring the sample tube between the M2020 rover and the Sample Return Lander. For these two spacecraft to successfully transfer tubes, the lander camera needs to localize and estimate the pose of the M2020 Bit Carousel with high accuracy. We present a method for fine pose estimation of the Bit Carousel, by using data from a simulated environment. We use a set of baseline simulated RGB-D images of the Bit Carousel and a camera matrix, to detect 2D features on the image and construct the corresponding 3D points on the dock. When a new test image is presented, we detect new features and match them to the baseline features. The matching results in a set of 2D-3D correspondences on the new image, from which we estimate the final pose. We evaluate our algorithm on a dataset of photorealistic simulated images, and examine the effect of varying the algorithm parameters. The results show that our method achieves an overall acceptable performance, with low pose errors and promising execution times.