Every year in the United States, approximately 2.5 million accidents are reported at intersections, as reported by the Federal Highway Commission[1]. The same agency reports that intersection accidents account for 40% of all crashes. Worse, 50% of all serious collisions and 20% of all fatal collisions occur at intersections.
A study conducted by the National Highway Traffic Safety Administration found “obstructed view” as a primary reason for intersection accidents across drivers of all ages and genders[2]. This will come as no surprise to any driver, though. Intersections pose difficulties that drivers do not experience on 2 lane roads, such as:
- Timed signals to monitor
- Cars from the side turning in front of driver
- Cars from in front turning in front of driver
- Cars stopping in front of driver to turn
- Pedestrians crossing in front of driver
- Pedestrians crossing beside driver
With all these reasons accounted for, it is no wonder how many accidents occur at intersections. In many cases, drivers simply have too many obstacles to account for at any given time. The proliferation of self driving cars may assist vehicles at intersections in accounting for a large number of obstacles simultaneously, but the problem of occluded viewpoints remains. With only car-mounted sensors, vehicles driving through intersections may not be able to detect obstacles such as crossing vehicles and pedestrians if their view is occluded by larger vehicles. The use of information gathered on sensors outside the vehicle will therefore be necessary to solve the occlusion problem.
To solve the occlusion problem, we are proposing an infrastructure system at intersections to monitor the area. This system will be able to detect obstacles (primarily pedestrians and vehicles) and predict their movements. Furthermore, the system will be designed to interface with vehicles approaching and passing through intersections, so that it can communicate with these vehicles. By utilizing path prediction and communication, the system will be able to detect would-be collisions in advance and alert vehicles, thus preventing collisions.
This project will develop a LiDAR/Camera-based system which can detect and track pedestrians and communicate with a single autonomous vehicle. The system will be tested on a controlled intersection with a known occluded space before being generalized to a four-way intersection with a traffic light. Our hope is that the technology we develop can be put to use in the future at all intersections, and ultimately for car-to-car sensor sharing.