Cross-spectral image alignment requires checking whether images observe the same objects, which capture times they represent and where a geometric model applies. Visible texture can differ from thermal-infrared response. Similar-looking grayscale values do not establish reliable correspondence. A normal visible-light calibration target is not necessarily identifiable in a thermal image.
Inspect overlap and identifiable targets
Record image resolution, lenses, cropping, mounting geometry and working distances. Select corresponding targets identifiable in both imaging modalities, including checks near image edges and occlusions. If a region is absent from one camera's view, software cannot recover its correspondence from nonexistent observations.
Establish timestamp meaning before comparing times
Distinguish device clocks, exposure-related times and host reception times. Confirm units, origins, restart behavior and clock synchronization. The manufacturer's timestamp documentation illustrates why the device type and its internal clock must be checked. Adjacent frames received by a host do not establish simultaneous exposures. Evaluate motion-related positional differences caused by timing differences separately.
Choose geometry for the scene
OpenCV calibration documentation distinguishes internal geometric parameters of a camera from geometry between cameras. A mapping estimated for an approximately planar scene requires checks of its distance and plane conditions. Objects at different depths with parallax should not inherit the same planar conclusion. Rectifying images does not automatically remove occlusion or establish correspondence for every pixel across imaging modalities.
Check applicability with independent observations
Keep observations used to estimate geometric parameters separate from independent checking observations. Record check-point locations, error definitions, distance, lighting and thermal conditions. Recheck parameter applicability after resolution, crop, lens or mounting changes. Image Alignment checks and temperature-measurement validation are separate. A visually aligned overlay is not evidence that temperature values are correct. This article reports no image-alignment-error or temperature-acceptance results for specified devices.
Related services: Vision algorithms and model engineering, project consultation.