Visual odometry involves the use of a sequence of images for navigation. Visual odometry relies on registering and optically following key points as they move across the sequence of images. The reliability of this procedure degrades when the images include dynamic objects or occlusions. This disclosure describes the use of generative artificial intelligence techniques to improve key point registration and visual odometry in severely occluded environments. Dynamic objects occluding a scene are identified and removed. The removed portions are filled in using generative artificial intelligence. In addition to conventional key points located on non-occluded sections of the image, synthetic key points are identified in the filled-in regions of the image. During navigation by visual odometry, the synthetic key points are used as secondary candidates in addition to the conventionally determined key points identified on non-occluded regions of the static environment.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.