Artificial Vision: A Key Component of Artificial Intelligence

In mid-June, while the Vivatech trade show in Paris brought together the elite of deeptech with a strong emphasis on artificial intelligence (AI), the Computer Vision and Pattern Recognition (CVPR) conference in Nashville, Tennessee, gathered over 7,000 academics and industry specialists in artificial vision (AV), which involves the automatic analysis of images and videos by a computer. This field is distinct from computer graphics, which focuses on image synthesis.

AV has been a cornerstone of AI since its inception. Without “seeing” the real world, there can be no interaction with it; without perception of the environment, there can be no robot, humanoid, drone, or autonomous car. Its origin is generally associated with Larry Roberts’ 1963 thesis, which focused on identifying very simple shapes (parallelepipeds and prisms) in images and estimating their position and orientation in space. This was historically the first example of a visual recognition system, which can be applied in various ways, such as determining whether a photo shows a cat or a dog, or whether a person is performing a specific action in a video.

Recognition is one of the two major axes of the field – the other being the three-dimensional reconstruction of a scene from multiple two-dimensional images (as the two eyes do when viewing a 3D film) – which are used in augmented reality applications or to allow a moving robot to avoid obstacles. Many other problems can also be addressed by AV, from the analysis of colors and textures to image restoration, and its applications range from industrial inspection to medical imaging and astrophysics.

Precision and Reliability

The geometric aspects of AV and depth perception dominated the 1990s, particularly with the pioneering work of Olivier Faugeras, with visual recognition long limited to very simple scenes.

---

Enjoyed this post by Thibault Helle? Subscribe for more insights and updates straight from the source.