Another option maybe to have a dumbed down sensor somewhere before the primary capture sensor which a.) exists at a known distance before the primary sensor. b.) continually scans the film c.) uses opencv to look for the edge of an exposure frame. When frame is picked up by this sensor and boundaries recognized, code tells it exactly when to trigger the primary scanner function based on the known distance between the two sensors and the known rotation of the motors.
Off the top of my head a potential problem with this could be frames that appear unexposed with not enough density to trigger the signal. Potential solution to this could be a pre-scan step that runs x amount of the current roll of film through the mechanism to baseline the frame positioning to how it is currently loaded on the machine and use that constant for the loaded roll.