iauh23/Dual-modality-target-detector ? reverse-engineered prompt

Build me a simple Python app that takes images from an RGB camera and a thermal camera, then combines the detections into one result.

I want to be able to click matching points in both images to calibrate them, calculate the transform between the two views, and then project the thermal detections onto the RGB image. For detection, keep it lightweight and use basic OpenCV image processing like thresholding, morphology, and contours rather than a big deep learning model.

After both cameras produce bounding boxes, merge overlapping ones so I end up with one clean set of detections using IoU based Non Maximum Suppression. Please make it easy to run with a small main script, and include the code needed to load the saved calibration points if they already exist. If you need to check current OpenCV docs while implementing anything, go ahead and do that.