Computer Vision News - May 2018

Computer Vision News Application 44 Application: AImotive AImotive is a global provider of AI- powered self-driving technology. They are a 180-people strong, investor-backed start-up, and growing. They just closed their third investment series and, compared to other start-ups, already have a significant income from their PoC project and prototyping project, with an increasing revenue year-on-year. There are many other big companies and start-ups around the world working on developing self-driving cars, but Gergely Debreczeni , Chief Scientist at AImotive, tells us that they are different because what they are developing is much more scalable. 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. They are fully committed to delivering Level 4 and Level 5 self-driving vehicles to the market in the near future. What makes their solution so scalable? Gergely says there are two things – cost- effective sensors and generic algorithms. Their hardware set up primarily uses cameras, and since so many cameras are produced each year for the mobile market they are low cost, but high performance and high resolution. The algorithms they are developing are generic – they are not specialising or overfitting algorithms for a specific scenario or geographical area. They are also agnostic to both use case and sensor. Gergely explains that their self-driving software stack has four modules – the recognition engine, location engine, motion engine, and control engine . The recognition engine is responsible for the recognition, detection, classification, and reconstruction of the environment around the vehicle. This part is based completely on deep learning and artificial intelligence. Gergely explains: “ The reason for this is very simple, we have found and we have proved that deep learning-based, neural network-based recognition classification algorithms are far more robust and far superior to anything else in this field. The

RkJQdWJsaXNoZXIy NTc3NzU=