A common approach in rendering engines to convert screen space coordinates to objects is to render a second image with light and shadow disabled where the color

A common approach in rendering engines to convert screen space coordinates to objects is to render a second image with light and shadow disabled where the color uniquely maps to an id. You then can uniquely identify 24 bits worth of objects without needing to maintain a KD tree.