The Vision Zero partnership is a crowdsourcing platform dedicated to help the video analytics system learn to detect road conflicts. The ADAS team at CVC, along with researchers from the UAB’s Engineering school, have joined the platform in order to generate realistic synthetic images with pixel-level annotation to help generalize and improve the training processes by deep learning algorithms.
More specifically, they will be using Synthia in order to enhance the accuracy of the machine learning systems under development in classifying objects into relevant categories – people walking, bicycling, or driving cars. As put by the partnership, “since the launch of the Video Analytics Towards Vision Zero crowdsourcing initiative over 500 people have volunteered to annotate existing traffic footage via the platform. The crowdsourcing initiative has enhanced the object classification accuracy of the machine learning system under development. People walking and bicycling are now accurately recognized 90% of the time when grouped together into a non-motorized object category”.
The partnership is coordinated by the Bellevue town Council and counts with different partners worldwide such as Microsoft, Unity, Lund University or ITS America. According to Jose Antonio Guitian, Cofund researcher at CVC and member of the ADAS team, “it’s an excellent opportunity to participate in a global consortium with shared interests and advancing in the improvement of ADAS research”.
The final aim of the partnership is to allow automatic systems and simulation to help prevent traffic crashes and save lives. The partnerships estimates that worldwide, 1.25 million people are killed annually in traffic accidents and that, with this new technology, “cities will be able to rapidly detect road conflicts and traffic engineers can then take preventative action to avoid crashes”.
Know more at the partnership’s website at: http://www.ite.org/visionzero/videoanalytics/