Computer Vision News - May 2018

45 Application: AImotive Computer Vision News Application recognition engine is purely deep learning-based. ” Gergely recalls some of the computer vision, hand-crafted, feature-based algorithms that they tried first in the recognition engine. For example, their first version of lane detection for the highway autopilot was based on classical computer vision algorithms . Those worked fine up to a certain level, but they could not be improved further on. Usability limits were reached very soon so they came to the conclusion that they should drop them. The motion engine is responsible for decision-making, path planning and project route planning of the vehicle. Gergely tells us: “ It’s a mixed solution using AI-based and classical algorithm- based solutions. The reason for this is that purely AI-based decision-making solutions cannot be verified very rigorously on the automotive level. For the motion engine, we need to combine the old classical solutions with the new AI-based ones in order to be flexible, human-like, and still safe. ” What’s next for AImotive? Gergely says that they have set up a development roadmap and are working through it. It consists of developing functionally- complete prototypes of automated parking, highway autopilot and city autopilot . They take projects from outside or from customers only if the project scope fits within their roadmap. Currently, automated parking is functionally complete - they have a working prototype and are just entering pre-production phase. The highway autopilot is a functionally-almost-complete conceptual prototype - it’s not yet in the pre- production phase but is progressing rapidly. City autopilot is in the research phase, so that’s at an early stage of development. Gergely concludes by saying how proud he is of the team: “ I think we managed to set up a team – my colleagues, the engineers, the complete personnel – and with this team we can do anything with high speed. I think that’s what I’m the most proud of and that’s one of the most important things that a company should create. ” This article was first published seven weeks ago on our magazine Bay Vision, spring 2018 “Their solution is not restricted to a specific use case or specific sensor, or limited by geographical area, so it is scalable across the globe.”