Fusing data from multiple vehicles into a common picture

Abstract

Modern day intelligent vehicles carry advanced sensors that gather data and fuse this into a map with dynamic objects. This overview of the car’s immediate surroundings is then utilized by an autonomous-drive pilot to steer the vehicle. There have been lots of research where the data collection and the data fusion virtually have been at the same location, on a single vehicle. However, less is known when the data providers and the consumer are separated, and especially when the providers are spatially spread. Through this contrasting approach, we envision increased awareness and safety for all involved actors. The separation introduces a communication problem, where the transmission time of the data becomes significant. In safety-critical systems, it is integral to receive new information as fast as possible. Thus, in order to study the dynamics of this model, we propose a novel test-track environment for Intelligent Traffic Systems (ITS) involving multiple vehicles where the data are centrally fused. The study is dissected into three areas: accuracy, network delay, and speed. Through this prototype system, we seek to illuminate difficulties and identify the possibilities of an on-line fusion server along with performance and scalability. Our tracking algorithm is based on Bayesian Tracking Theory and tested with data provided by Volvo Car Corporation. Using this setup, our simulations show significant improvements in positional accuracy when the vehicles cooperate. We also find that the transmission delay is crucial in terms of both speed and precision of the algorithm. In addition, the bottlenecks were primarily located to algebraic operations, which put a boundary of the speed of our tracking algorithm. Due to the small scale of the system and the matrices, the possibility of optimizations was done by multithreading. The study made on our fusion server suggest that an on-line fusion server might support a few hundred cars. But what dictates the scalability is the trade-off between the speed and precision of the fusion server - which is extremely difficult, if not impossible to balance optimally.