Cadillac’s parent, GM, is proposing a camera view that recolors selected road signs to help drivers with color vision deficiencies distinguish them. A patent filing discovered by Carmoses describes software that cuts a small patch from an image, adjusts its colors and puts it back. Most of the scene stays untouched, reducing the amount of image data that needs the treatment.
GM has previously described personalizing vehicle display colors using a driver’s responses to color tests. This approach combines localized processing with an optional profile that can tailor the result to its viewer. Its usefulness also depends on how quickly the system can deliver the adjusted picture.
Nothing in the filing names a specific model. We used the Cadillac Lyriq for these renders because its 33-inch curved display suits a feature built around camera imagery. The interface graphics show how the proposed functions might look, not an announced Cadillac menu.
A Small Patch Does The Work
Vehicle cameras capture the surroundings, and a detector searches the image for targets. GM specifically includes stop signs, street signs and directional signs, while allowing other objects. Once something is identified, the software defines a box around it and adds a preset margin measured in pixels.
The adjustment covers the sign and some nearby scenery, rather than following the object’s outline precisely. The selected rectangle is cropped, processed and inserted into the same location in the original picture. The completed view then appears on an in-car display.
The driver receives the altered image without having to identify each sign manually. GM does not specify how the feature is switched on or where its camera view appears. The proposal operates on displayed imagery, so the road sign outside retains its original appearance.
One processing route uses daltonization, a technique that adjusts image colors to make them more distinguishable to people with color vision deficiencies. Here, that adjustment runs only on the selected patch. The approach does not require a generative AI model, although GM also describes one.
The Same Sign, Different Palettes
The alternative uses a generative adversarial network, or GAN, trained on paired original and color-adjusted images. One part learns to produce the translated image, while another evaluates the translation. The trained system then applies its learned color adjustment to the cropped region.
This version can use a profile containing the individual’s color-vision data. Instead of relying solely on a standard transformation, it can customize the output using information about the person viewing it. Stored profiles may be updated through user input or repeated operation of the software.
The filing does not specify the replacement colors or explain how a driver’s color-vision data would initially be collected. Nor does it supply testing that establishes which adjustments work for which users.
Recoloring also depends on finding the right object in the first place. GM does not explain what the display does when detection fails, or how it handles an incorrect identification.
Recognition Matters More Than Processing Speed
GM says working on selected regions produces results at approximately 1.5 times the speed of transforming the entire image. It supplies no benchmark setup, measured delay or specified test hardware. That leaves the size of the practical improvement unresolved, especially once object detection and reinserting the patch are considered.
Hyundai and Kia have proposed changing the colors of moving objects in surrounding-camera images for drivers with impaired color vision, including views from digital side mirrors. Their approach can recolor an object’s border or its entire area.
Cadillac already offers a Rear Camera Mirror on certain Lyriq trims, giving this proposal a familiar context for buyers accustomed to viewing their surroundings electronically. GM’s proposed software adds a different objective to a camera display, changing selected colors to suit the person using it.
GM’s targeted approach is a sensible use of image processing, and the option to adapt it to an individual is its strongest feature. The decisive test should be whether drivers recognize relevant objects more reliably and promptly. Faster color processing earns its place on the dashboard when it helps the driver recognize a relevant road sign sooner.
