Computer Vision-Based V2X Collaborative Perception

This paper presents the computer vision-based V2X collaborative perception. Our system uses a forward-looking camera in the host vehicle. The camera detects road users such as pedestrians, vehicles, and motorcycles. Such information includes object type, relative location, direction, and speed. This information is used to compose proxy Basic Safety Messages on behalf of the detected objects. Early adopters of the V2X technology can experience the benefits of enhanced V2X market penetration. The outcome of adopting this concept will result in an inflated V2X market penetration rate leading to earlier safety, mobility, and situational awareness improvements. The ultimate goal is for all road participants to be fully aware of each other. The novelty of our work is the integration of computer vision-based detection and LTE-V V2X communications, in addition to implementing the concept for pedestrians and bicyclists. A similar concept will work for other sensors such as radar and LiDARs with varying detection and classification capabilities. This work show that our concept is both feasible and beneficial. For position accuracy measurement, we compare the local GNSS and inferred localization of the detected road users. The average localization difference ranges from 1-3m in general. This paper presents the method and the concept, followed by the system description and analysis of the results.