Object localization in an image.

I have designed a small robotic arm and I would like to implement a computer vision system using image processing to localize an object and pick it. The idea is that a camera is mounted from above and a shot of the workspace is taken at first without any object as a reference and then another picture with the object in place. The algorithm is based on having the reference image and the working image to be the same (theoretically) and the only difference would be the object introduced in the workspace. Then the relative position is calculated to get the coordinates of the object relative to the robot. The robot's location is already determined by template matching but I would like to have ideas about an efficient way to determine the object's location knowing the it could have any shape and it could be a random object without any training or machine learning whatsoever based only on image processing tweaks.

I have subtracted both images and done some post-processing but it still requires some work. The problem is to deterine this object global location.

PS: Disregard the lubrication stains on the sheets.