The main idea behind the initiative is to make the solution fast, accurate and self-learning whenever possible. The final goal is to get the system constantly improving in the background in terms of semantic connections and complex scenes analysis while performing accurate segmentation.

A pretty straightforward theory of evolution will be used to illustrate the level of "intelligence".

Traditional solutions are based on single-image/frame segmentation, which introduces certain limitations as well as an inability to capture various important details. Like, for example, temporary overlapping objects, dynamic light conditions, etc.

Multi-cam vision uses streams from several cameras to achieve more accurate and robust scene segmentation. It is a common approach to combine segmentation pipelines with domain knowledge to make segmentation results better.