Towards reliable deployment of computer vision systems in the real world

Vehicles with autonomous parking, lane detection and distance estimation are common in new cars and of great convenience and safety for drivers and passengers. However, autonomous systems in cars are usually trained' in common and static environments, so they can malfunction in unusual environmental conditions (e.g., extreme smoke or rain). Car manufacturers need a solution to this safety risk and that solution lies in adaptive autonomous systems that can adapt to any road conditions. This project will design a novel, world-first simulator and algorithm that minimises the gap between real world conditions and those in the simulated environment and adaptively select the best systems to create a more reliable algorithm for the real world. The computer program we create can be retrofitted to existing cars and embedded in all new cars. Working with our long-standing industry partner,