It is crucial for autonomous road vehicles to accurately perceive and understand traffic signs in the real world. To teach this skill, many efforts have been made to create traffic sign datasets for training deep-learning algorithms. However, datasets have been limited in their scale or diversity, until now. Available for both commercial and research purposes, Mapillary has used computer vision to automate and scale mapping and launched what it calls the world’s largest and most diverse publicly available traffic sign recognition dataset to teach autonomous vehicles to understand traffic signs. The Mapillary Traffic Sign Dataset consists of images from around the world featuring high variability—from weather and times of day to different camera sensors and viewpoints. Although traffic sign recognition technology is fairly common, the company says this is the first time such a large and diverse dataset has been launched for anyone to license for training their own traffic sign recognition systems. A collaborative approach means that all 570 million images on the Mapillary platform have been uploaded by people and companies from all over the world. About 100,000 of these images were selected for the Traffic Sign Dataset. More than 300 different traffic-sign classes have been verified and annotated, resulting in more than 320,000 labeled traffic signs across the images. Over 52,000 images have been fully verified and annotated by humans, with the remaining images annotated partially through Mapillary’s computer vision technology. This new offering comes months after research showed that inexpensive cameras have the potential to catch up with LiDAR in teaching autonomous vehicles to understand their surroundings, something that could reduce the cost of an autonomous vehicle by tens of thousands of dollars. Mapillary’s new dataset tackles the perception part of the problem, but diversity in the dataset is key in moving toward camera-based solutions.

Click to view AVT’s 2020 ACES Award Winners best in class of