Car-to-Pedestrian communication with MEC-support for adaptive safety of Vulnerable Road Users

Abstract

Car-to-Pedestrian (Car2P) communication has a huge potential for preventing potential accidents between pedestrians and vehicles by exchanging context information. Similar communication principles have been explored for Car-to-Car (C2C) communication since quite a while and are now being deployed as an additional safety mechanism. Unfortunately, Vulnerable Road Users (VRUs) such as pedestrians and bicyclists are not yet covered by these systems. Looking at pedestrians, we explore possibilities for the exchange of relevant safety mechanisms between pedestrians and cars making use of readily available communication capabilities of current LTE networks. Necessary activity and collision detection algorithms have to be run on the users’ smartphones in order to determine the criticality of the situation. In order to improve both the overall system latency and the energy efficiency of the smartphone applications, we suggest the use of Multi-access Edge Computing (MEC). We implemented the system to perform measurements on the smartphone side as well as extensive simulations on the networking side. Our results clearly show the advantages of our concept and the integrated MEC approach.

Introduction

In recent years, cooperative collision detection has emerged as a complementary approach to traditional safety solutions using vehicle sensors like cameras, laser scanner, and RADAR to detect nearby objects. Different from the sensor-based approach, which requires a direct Line-Of-Sight (LOS) between vehicles and detected objects, cooperative collision detection can even operate in Non-Line-Of-Sight (NLOS) scenarios by exchanging movement information between entities via wireless communication. In the context of safety systems for Vulnerable Road Users (VRUs) like pedestrians and bicyclists, vehicles exchange information with VRUs’ mobile smart devices, e.g., smartphones, tablets, or wearable devices [1], [2], [3], [4].

The general concept of a VRU safety system is shown in Fig. 1 (extending our earlier work in [5]), in which vehicles and VRUs have two possible options for data transmission, namely direct Device-to-Device (D2D) communication and an infrastructure-based approach. As an initial idea, D2D technologies, such as Wi-Fi Peer-to-Peer (P2P) or Dedicated Short Range Communication (DSRC), are designed to serve the need of cooperative safety systems, which allows messages to be directly transferred from sender to receiver. In particular, DSRC has been assumed for Car-to-Car (C2C)/Vehicle-to-Vehicle (V2V) and Car-to-Everything (C2X)/Vehicle-to-Everything (V2X) communication and is likely to be available for smartphones in the next few years [6], [7]. However, these D2D technologies do not offer good coverage and range, i.e., less than 200 m for Wi-Fi Direct and less than 1000 m for DSRC, especially around intersections in urban areas [8], [9]. To overcome these limitations, cellular technology is proposed as an alternative.

Cellular V2X (C-V2X), or more specific LTE-V2X, includes two interfaces: the LTE interface (named Uu), that supports the communication between vehicles/end-devices and mobile base stations and the newer D2D interface (named PC5), that enables V2V communications based on direct LTE sidelink [10], [11]. C-V2X is indicated to have many advantages in different aspects, such as communication range, performance, and reliability, especially with the current evolution from LTE to 5G [12]. PC5, like DSRC, has been studied by several existing works and is deployed mainly for vehicles [10], [11], [13]. This direct V2X communication can significantly reduce the latency and is expected to become superior to DSRC [12]. Meanwhile, the Uu radio interface, which is already available for smartphones, still remains an important part of the next generation wireless V2X system with the ability to support long-range communication. However, the feasibility of applying this interface and network infrastructure to provide an additional approach for different V2X use cases is still unclear. Recognizing this gap along with the desire to leverage existing networks for V2X communications, we focus on studying the performance of cellular-based systems over the Uu interface and exploiting this approach for VRU safety applications/services. To be more specific, Car-to-Pedestrian (Car2P), a typical case of VRU systems, in which the safety for pedestrians is the main objective of our investigation in this paper.

In order to enable collision detection, pedestrians’ smartphones and vehicles need to send beacon messages periodically to each other and after obtaining the beacons, a Collision Detection Algorithm (CDA) is executed. While these tasks are not a big deal for vehicles, they could be a significant obstacle when running on smartphones with limited resources. Especially, it would be a heavy burden on smartphones in terms of energy consumption and executing time when more information from other vehicles is needed and a more complex CDA is used. In this paper, we concentrate on addressing this bottleneck by considering the use of computation offloading using Multi-access Edge Computing (MEC). Applying this mechanism for smartphone applications, such as gaming, image/video processing, object/face recognition, or accelerated web browser has been shown to have a noticeable improvement in energy efficiency [14]; however, its applicability in the context of Car2P safety systems is still unclear.

The original idea of our approach was presented in our recent work [15], in which we proposed a dynamic selection of execution location (local or remote) for calculating pedestrian context information. Conceptually, the beacons transmitted between objects comprise some fundamental information like current position, heading direction, and speed [16], [17], [18], [19]. Some recent works [3], [20] suggested using additional context information to improve the accuracy of prediction results. Looking in more details, we can see that it is quite simple to calculate some information like position or direction directly from smartphone sensor data [6], [21]. But for others, e.g., motion states or activity, more extensive data handling like pre-processing, extraction of features, and training machine learning models are needed [20], [22], [23]. Depending on the amount of raw sensor data collected, an appropriate computational scheme is selected to save energy on smartphones.

In this paper, we extend and improve our proposed approach in [15], to support not only context information calculation but also the CDA. Generally, in cellular-based safety systems, context information computation and collision detection could be done either on a pedestrian smartphone itself or on a remote server. Deciding the strategy to not only prolong the battery life of the smartphone but also to meet the timing constraints in a Car2P safety system is a challenging task. It requires to take into consideration different parameters, such as energy consumption and processing time of local computation, network communication overhead, as well as real-time requirements of the system.

To provide insights on this adaptive approach, we carry out experimental and simulation studies on the performance of a Car2P system when applying both Local and Offload computational schemes on smartphones. In addition, energy consumption and latency of each scheme are investigated in the relationships with different parameters, such as the machine learning algorithm, the sensor sampling rate, the window size, and the sending interval of messages on the smartphone. We see the results of this paper as an important step towards energy-efficient VRU safety systems.

Our main contributions can be summarized as follows:

• We propose an adaptive computational approach for smartphones, which considers the possibility of offloading tasks in both data and service levels of Car2P safe applications.

• We measure and analyze the energy consumption and processing time of a lightweight machine learning application for determining pedestrian activities as well as a CDA for detecting potential collisions.

• We simulate a scenario of a Car2P system using the simulation framework Veins LTE to evaluate the end-to-end performance and scalability of our adaptive approach.