Vehicular networking and computer vision-based distance estimation for VANET application using Raspberry Pi 3

Abstract

This research was implementing vehicle networking using WIFI connection and computer vision to measure the distance of vehicles in front of a driver. In particular, this works aimed to improve a safe driving environment thus supporting the current technology concept being developed for inter-vehicular networking, VANET, especially in its safety application such as Overtaking Assistance System. Moreover, it can wirelessly share useful visual information such as hazard area of a road accident. In accordance with Vehicle-to-Vehicle (V2V) concept, a vehicle required to be able to conduct networking via a wireless connection. Useful data and video were the objects to be sent over the network established. The distance of a vehicle to other vehicles towards it is measured and sent via WIFI together with a video stream of the scenery experienced by the front vehicle. Haar Cascade Classifier is chosen to perform the detection. For distance estimation, at least three methods have been compared in this research and found evidence that, for measuring 5 meters, the iterative methods shows 5.80. This method performs well up to 15 meters. For measuring 20 meters, P3P method shows a better result with only 0.71 meters to the ground truth. To provide a physical implementation for both the detection and distance estimation mechanism, those methods were applied in a compact small-sized vehicle-friendly computer device the Raspberry Pi. The performance of the built system then analyzed in terms of streaming latency and accuracy of distance estimation and shows a good result in measuring distance up to 20 meters.