Predicting Trajectory Paths For Collision Avoidance Systems

• Main Author: Barrios, Cesar
• Format: Others
• Language: en
• Published: ScholarWorks @ UVM 2015
• Subjects: Collision; GIS; Kalman; V2I; V2V; Electrical and Electronics
• Online Access: http://scholarworks.uvm.edu/graddis/386; http://scholarworks.uvm.edu/cgi/viewcontent.cgi?article=1385&context=graddis

This work was motivated by the idea of developing a more encompassing collision avoidance system that supported vehicle to vehicle (V2V) and vehicle to infrastructure (V2I) communications. Current systems are mostly based on line of sight sensors that are used to prevent a collision, but these systems would prevent even more accidents if they could detect possible collisions before both vehicles were in line of sight. For this research we concentrated mostly on the aspect of improving the prediction of a vehicle's future trajectory, particularly on non-straight paths. Having an accurate prediction of where the vehicle is heading is crucial for the system to reliably determine possible path intersections of more than one vehicle at the same time. We first evaluated the benefits of merging Global Positioning System (GPS) data with the Geographical Information System (GIS) data to correct improbable predicted positions. We then created a new algorithm called the Dead Reckoning with Dynamic Errors (DRWDE) sensor fusion, which can predict future positions at the rate of its fastest sensor, while improving the handling of accumulated error while some of the sensors are offline for a given period of time. The last part of out research consisted in the evaluation of the use of smartphones' built-in sensors to predict a vehicle's trajectory, as a possible intermediate solution for a vehicle to vehicle (V2V) and vehicle to infrastructure (V2I) communications, until all vehicles have all the necessary sensors and communication infrastructure to fully populate this new system. For the first part of our research, the actual experimental results validated our proposed system, which reduced the position prediction errors during curves to around half of what it would be without the use of GIS data for prediction corrections. The next improvement we worked on was the ability to handle change in noise, depending on unavailable sensor measurements, permitting a flexibility to use any type of sensor and still have the system run at the fastest frequency available. Compared to a more common KF implementation that run at the rate of its slowest sensor (1Hz in our setup), our experimental results showed that our DRWDE (running at 10Hz) yielded more accurate predictions (25-50% improvement) during abrupt changes in the heading of the vehicle. The last part of our research showed that, comparing to results obtained with the vehicle-mounted sensors, some smartphones yield similar prediction errors and can be used to predict a future position.