Automatic Building and Labeling of HD Maps with Deep Learning

In a world where autonomous driving cars are becoming increasingly more common, creating an adequate infrastructure for this new technology is essential. This includes building and labeling high-definition (HD) maps accurately and efficiently. Today, the process of creating HD maps requires a lot of human input, which takes time and is prone to errors. In this paper, we propose a novel method capable of generating labelled HD maps from raw sensor data. We implemented and tested our methods on several urban scenarios using data collected from our test vehicle. The results show that the proposed deep learning based method can produce highly accurate HD maps. This approach speeds up the process of building and labeling HD maps, which can make meaningful contribution to the deployment of autonomous vehicles. I -Introduction During the last couple of decades, autonomous driving has become an important research topic in the scientific community. This stems from the fact that scientists, governments and people in general are starting to realise the huge positive impacts that autonomous driving could have on our daily lives. According to a report by The Department of Transportation of the USA, self-driving cars could reduce traffic fatalities by up to 94% by eliminating the accidents that are due to human errors. The race toward fully autonomous driving cars, or level 5 autonomy as categorized by SAE International, has given rise to multiple new fields and was the catalyst to launch or greatly improve several new disciplines. This manifested itself in the field of deep learning, which has made it possible today to achieve a respectable level of autonomy when driving on roads that fall into the classic scenario box. Lanes and roads (or driveable regions) detection can be achieved with neural networks trained to excel in task related to pixel-wise segmentation of images captured by cameras (Long, Shelhamer, and Darrell 2015), making the car aware of where it is safe to drive. Other types of networks are trained to detect obstacles and classify them into several independent classes (Redmon and Farhadi 2017), sometimes with the help other