Risk Maps Based on Vision and 3D LiDAR

Traffic accidents are a major concern in the current mobility context, both for driverless and human driven vehicles. In this line, the early assessment of collision risk is paramount in order to predict and avoid accidents and casualties involving the targets in the road. For that purpose, this paper proposes a solution of creating continuously updated risk maps around the ego vehicle, along with some risk indicators to assess the risk of collision before it occurs. Those maps are based on the properties of the targets perceived by the sensors onboard the vehicle, along with models of motion and occupancy prediction. This domain of risk prediction and assessment is very vast, and this paper focuses on combining visual perception for the classification of targets, and 3D LiDAR to extract the relative location, dimensions and motion properties of the targets. This information is fed into probabilistic occupancy maps and risk maps are created at the rate of sensor data stream, both as visual information for humans, but also as risk descriptors to trigger alert and emergency actions. The results in simulation and in real scenarios show consistent and valid information that can be further used in real scenarios on the road.