Auriga the Tool of a marking of video and stream of data for machine learning

Along with permission of difficult moral dilemmas, fast and exact object recognition of the world around is one of the most difficult tasks for creators of unmanned vehicles. The majority of autonomous vehicles use a combination of touch technologies that "see" the road. Sensors of detection of distance, such as lasers and radars, report distance to the objects surrounding the vehicle. Visual sensors, such as cameras, will recognize color and parts of a landscape. Many producers of unmanned vehicles developed the systems of deep learning which learn to drive safely the car in different conditions, based on a huge number of the marked data from sensors.

Within the big project on control automation by the Auriga car develops the semi-automated tool for a marking of video and formation of the stream of data for machine learning. The user will be able to load into the application video data (and also data from radars and data on speed) and "teach" to distinguish and mark automatically the application objects according to the set set of labels – for example, a road marking, traffic lights and road signs, trees, other machines, cyclists, pedestrians, etc. The marked data create the stream of data for the subsequent machine learning and control automation by the car. The more data it will be marked, the system will be better "see" the road in the future.

As cameras are very susceptible to weather conditions, it is important to consider change of day and night, feature of seasons, a rain, fog, a mist and also snow which can hide a road marking and even signs completely. Unmanned vehicles rely on well-defined rules of traffic therefore in extreme conditions it is difficult for them "understand" what occurs around. All these circumstances were considered by our engineers: driving videos in adverse weather conditions make a considerable part of the created lake of data.

Robotics