Location Estimation and Trajectory Prediction for Collision Risk Assessment in Connected Vehicle Environment
Date
2019-09-18
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Safety systems in intelligent and autonomous vehicles rely heavily on the accuracy of localization and location prediction of nearby road users. Current vehicular systems use a variety of sensors to perceive the environment. Cameras, proximity and ranging sensors are the most common types of devices used for this purpose. The main limitation of onboard sensors is the partial perception of the surrounding environment due to occlusions, limited field of view, or resolution and range restrictions. Wireless vehicular communication offers new opportunities for safety applications via information sharing and extending the perception of a car, beyond the limitations of its onboard sensors. This thesis first explores the problem of fusing multiple sources of location information, including the sensor data and information shared via Vehicle-to-Vehicle (V2V) communication to improve localization accuracy. Using sensor data adds more challenges as it is usually noisy, mixed with clutter and false alarms. To address these challenges, the problem is formulated in Random Finite Set (RFS) statistics and solved via the Probability Hypothesis Density (PHD) filter. Second, this thesis investigates the location prediction problem in the connected vehicle environment. A data-driven framework is proposed to learn motion patterns from historical trajectory data via Gaussian Process Regression (GPR) and share this information among vehicles. In this framework, a vehicle leverages GPR models alongside sensory location data to predict the positions of nearby cars. Third, to improve the accuracy of both location estimation and prediction, a new multi-target Bayesian filtering algorithm is proposed that incorporates the GPR models in the Multi-source Multi-target Bayesian filters. Simulations based on real-world data and comparisons to similar algorithms from the state-of-the-art demonstrate the performance of the proposed methods. The results show about 30% improvement in estimating and predicting the location of surrounding vehicles for seconds in advance, fulfilling the requirements for a real-time collision risk assessment system.
Description
Keywords
Connected Vehicle Environment, Bayesian Filtering, Collision Avoidance, Intelligent Vehicle, Machine Learning
Citation
Afkhami Goli, S. (2019). Location Estimation and Trajectory Prediction for Collision Risk Assessment in Connected Vehicle Environment (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.