Computer Vision-Assisted Instant Alerts in 5G

This paper introduces an innovative model which incorporates vehicle On-Board Unit (OBU) data and roadside video information to provide instant alert messages to drivers. We apply computer vision techniques to perform real-time danger event detection and to identify specific surrounding vehicles that should be alerted. Different from traditional broadcast-based alerting, we propose to send these instant alert messages to the target vehicles by unicast and geocast. To do so, an accurate method is required to analyze the spatial relation of vehicles. Also, to confine our alert messages to only those target vehicles, we rely on roadside cameras and apply a sensor fusion technique that can link a video object with its communication MAC address. Through this innovative idea, we integrate computer vision with 5G networks and enable transmitting instant alerts to precise vehicles without interfering irrelevant vehicles. How to incorporate our system with 3GPP V2X by setting proper transmission parameters is also addressed. To validate our idea, we present four common road danger events and show how our model works. To the best of our knowledge, this is the first work bringing computer vision to instant messaging.