One possibility (what we did last year) - in your image processing code, push the information about rectangles (position, size) into a
NetworkTable. Then,
write a SmartDashboard widget with a transparent background that reads this information and draws rectangles. Manually drag this overlay to align with the camera feed.