Cyclist Speed Estimation Using Accelerometer, Gyroscope, and Monocular Camera

V2X communication and its supporting systems are significant for the future of fully autonomous driving systems because onboard sensing may suffer from its limited detection range or physical obstruction. Cyclists are more dynamic compared to pedestrians, and their lack of signals to show their intentions of crossing and turning to autonomous vehicles may lead to traffic accidents. This paper proposes to estimate cyclists' speed using data from portable devices such as smartphones to support V2X communication systems for autonomous driving. Prior research shows that the data gathered from a smartphone's accelerometer and gyroscope allow the model to estimate a cyclist's speed that surpasses the GPS-based speed estimation method. However, the results of the accelerometer and gyroscope-based method were noisy. On the other hand, visual sensors have been widely used for visual odometry. Monocular visual odometry may encounter issues caused by scale drift and dynamic objects. This paper aims to integrate three sensors, an accelerometer, a gyroscope, and a monocular camera, into a Machine Learning model to improve the performance of cyclist speed estimation. A series of experiments have been conducted to demonstrate the effectiveness of the proposed method.

I. Introduction

As a vehicle, bicycles have found their way to be one of the most popular means of transportation. By the year 2000, a survey by Koike et al. [1] showed that bicycles in Japan numbered more than 80 million. In 2021 alone, 359 people were killed in accidents related to bicycles as stated by the Japanese National Police Agency [2]. The direness of this problem has only become even more apparent as companies and organizations of different industries are competing on creating the best autonomous driving technology. They mostly share the same objective of manufacturing automated vehicles that can fully depend on their onboard sensing and intelligence. These vehicles, however, will most likely be hindered should their capacity to detect objects are physically obstructed. Thus, developing a road user safety system utilizing V2X communication is a bare necessity as this allows those connected to the system to relay necessary information from cyclists to an automated vehicle.