Map Points Between Image Coordinates
A point on one conveyor image must be located on another view.
Read the full lesson and open its transcript below. A narrated video will be added after production.
What you will learn
A point on one conveyor image must be located on another view.
A practical question
Two cameras watch the same conveyor, and a scratch seen in one image needs to be found in the other. Its x and y numbers cannot simply be copied. Each camera has its own image coordinate frame, with its own origin, scale, and viewpoint.
The useful idea
We need a measured mapping from points in the first frame to points in the second. Name the source and destination every time; reversing a mapping can send a perfectly valid point somewhere wrong. A coordinate method using homogeneous coordinates lets us express translations alongside other transformations. That is bookkeeping with a purpose: it keeps the full mapping in one operation.
Work through the example
Mark a visible corner of the product in both images. Estimate the mapping from several known correspondences, then transform another corner and see where it lands. If the two cameras see a flat conveyor from different perspectives, we may eventually need a projective mapping rather than a simple rotation or shift. A coordinate is tied to one image grid. If the camera view shifts or a map is warped, an annotation copied without its transformation points to the wrong place. Pick a recognizable corner, write down its original horizontal and vertical coordinates, and apply the measured mapping. Then look for that corner in the second image. Test several points across the field; agreement near one corner does not guarantee accuracy everywhere. This habit prevents a plausible box from silently drifting off the object. Now check the idea on a concrete example: Write a point as (x, y) in one image, then apply the measured mapping to compare the second. A possible trap is worth naming: Mixing row-column with x-y order creates wrong but numerically valid coordinates. The answer follows from that evidence: Coordinate transforms map positions between frames with a defined origin and units.
Limits and next step
Record which image defines each coordinate frame and inspect points overlaid on the second camera's original view. Next, we'll handle a useful simpler family of transformations: affine warps.
Worked example
Write a point as (x, y) in one image, then apply the measured mapping to compare the second.
Quick check: test your understanding
Try it yourself: Consider this case: Write a point as (x, y) in one image, then apply the measured mapping to compare the second. Explain how you would check the result. In your answer, address this limitation: Mixing row-column with x-y order creates wrong but numerically valid coordinates.
Check the reasoning
Coordinate transforms map positions between frames with a defined origin and units. Mixing row-column with x-y order creates wrong but numerically valid coordinates.
What is the main idea behind map points between image coordinates?
Coordinate transforms map positions between frames with a defined origin and units.
How is it used in the example?
Write a point as (x, y) in one image, then apply the measured mapping to compare the second.
What limitation should be checked?
Mixing row-column with x-y order creates wrong but numerically valid coordinates.
Further reading
Digital Image Processing, 4th edition — Rafael C. Gonzalez and Richard E. Woods.
Original ToolsInVision teaching examples. Editorial approach · Suggest a correction
Accessible referenceLesson 49 transcript327 words · Narration ready for the upcoming videoOpen transcript
Two cameras watch the same conveyor, and a scratch seen in one image needs to be found in the other. Its x and y numbers cannot simply be copied. Each camera has its own image coordinate frame, with its own origin, scale, and viewpoint.
We need a measured mapping from points in the first frame to points in the second. Name the source and destination every time; reversing a mapping can send a perfectly valid point somewhere wrong. A coordinate method using homogeneous coordinates lets us express translations alongside other transformations. That is bookkeeping with a purpose: it keeps the full mapping in one operation.
Mark a visible corner of the product in both images. Estimate the mapping from several known correspondences, then transform another corner and see where it lands. If the two cameras see a flat conveyor from different perspectives, we may eventually need a projective mapping rather than a simple rotation or shift. A coordinate is tied to one image grid. If the camera view shifts or a map is warped, an annotation copied without its transformation points to the wrong place. Pick a recognizable corner, write down its original horizontal and vertical coordinates, and apply the measured mapping. Then look for that corner in the second image. Test several points across the field; agreement near one corner does not guarantee accuracy everywhere. This habit prevents a plausible box from silently drifting off the object. Now check the idea on a concrete example: Write a point as (x, y) in one image, then apply the measured mapping to compare the second. A possible trap is worth naming: Mixing row-column with x-y order creates wrong but numerically valid coordinates. The answer follows from that evidence: Coordinate transforms map positions between frames with a defined origin and units.
Record which image defines each coordinate frame and inspect points overlaid on the second camera's original view. Next, we'll handle a useful simpler family of transformations: affine warps.