Data Collection and Processing Methods for the Evaluation of Vehicle Road Departure Detection Systems .pdf

Libria

231

6页

0次

2021-02-22

50墨值下载

Data Collection and Processing Methods for the Evaluation of

Vehicle Road Departure Detection Systems

Dan Shen, Qiang Yi, Lingxi Li, Stanley Chien, Yaobin Chen,

Transportation Active Safety Institute

Indiana University-Purdue University Indianapolis, USA

Rini Sherony

Collaborative Safety Research Center (CSRC),

Toyota Motor North America, USA

Abstract— Road departure detection systems (RDDSs) for

avoiding/mitigating road departure crashes have been

developed and included on some production vehicles in

recent years. In order to support and provide a

standardized and objective performance evaluation of

RDDSs, this paper describes the development of the data

acquisition and data post-processing systems for testing

RDDSs. Seven parameters are used to describe road

departure test scenarios. The overall structure and

specific components of data collection system and data

post-processing system for evaluating vehicle RDDSs is

devised and presented. Experimental results showed

sensing system and data post-processing system could

capture all needed signals and display vehicle motion

profile from the testing vehicle accurately. Test track

testing under different scenarios demonstrates the

effective operations of the proposed data collection

system.

I. INTRODUCTION

Vehicle crashes due to road departure is a leading cause

of fatalities on US highways [1]. Approximately 12,000

drivers lose their lives each year due to road departure crashes

[1]. Roadside crashes account for about 35 percent of the

fatalities on nation’s highways [2]. Road departure warning

and road keeping assistance (RKA) systems are active safety

technologies for dealing with this problem. Most of currently

developed lane/road departure mitigation systems are based

on the detection of lane markings. In addition, road departure

detection systems mostly work on straight roads and slightly

curved roads due to technological difficulties of road edge

detection. However, many roads do not have lane markings

or clear lane markings, especially in some rural and

residential areas. Therefore, road departure detection and

avoidance technologies could rely on the detection and

identification of different types of road edges and road

boundary objects on any kinds of road geometries.

Safety professionals and automobile manufacturers have

strived to overcome the road departure issue by developing

modeling methods, perception algorithms, and control

strategies for road departure warning/mitigation systems. A

simulated road departure detection system that relies on

roadside terrain geometry analysis and subsequent threat

assessment was discussed in [3]. The viability of detecting

vehicle run-off the road through the measurements of

anomalies under scenarios that left and right tires experience

force imbalance was investigated in [4]. Authors in [5]

explored effectiveness of a three-layer perceptron neural

network to predict an unintentional road departure. A driving

simulator testing that evaluates the road-departure prevention

system in an emergency was presented in [6]. Authors in [7]

proposed a system based on a closed-loop driver decision

estimator (DDE), which determines the risk of road departure.

Some other related research works include traffic forecast

using deep learning method [8], evaluation of lane departure

correction system (LDCS) based on the stochastic driver

model [9], analysis of the LDCS utilizing naturalistic driving

data [10], and so on.

Recently, more and more active safety technologies have

been studied and developed on vehicle electronic stability

control systems for advanced RDDSs. However, a

fundamental question remains to be answered: “how to

evaluate and demonstrate the effectiveness of the vehicle

RDDSs on real roads?” Many testing and verification

methods have been proposed based on virtual modeling and

computer simulations [11-14]. However, comprehensive

testing which combines virtual simulations and real-world

testing is crucial. Therefore, the development of testing

methods for vehicle road departure detection system is

important. One such method is testing on a test track.

RDDSs on the market are only used on straight roads and

slightly curved roads with clear lane markings. The primary

goal and main contribution of this work is the development

of a comprehensive method for testing vehicle road departure

mitigation systems and for evaluating the road edge detection

and control efficiency of RDDSs on all types of roads

with/without lane markings. This approach includes test

equipment development, and data collection and processing

procedure, which provides a practical guidance on the

development of next generation intelligent RDDSs with

consideration.

The remainder of this paper is organized as follows. The

key variables of road departure test scenarios based on

2018 IEEE Intelligent Vehicles Symposium (IV)

Changshu, Suzhou, China, June 26-30, 2018

representative national crash databases are discussed and

determined in Section II. The overall structure for testing the

vehicle road departure detection system is presented in

Section III. Section IV describes the structure of data

collection systems. Data post-processing system is presented

in Section V. Finally, the conclusions and comments on the

performance of the proposed method are given in Section VI.

II. KEY PARAMETERS IN ROAD DEPARTURE TESTING

One question to be answered in road-departure mitigation

system evaluation is what the representative testing conditions

are. This question is addressed in this paper by focusing on the

information and key parameters regarding the representative

test scenarios and the vehicle parameter (departure speed vs.

departure angle). The overall structure of road-departure

evaluation system to be discussed in Section III is also based

on the outcomes of key variables of road departure tests.

Since all test methods and results rely on road-departure

test scenarios, the determination of key variables in these

scenarios from national crash databases is crucial. To achieve

this objective, an overall approach and process for road

departure testing is proposed, as shown in Fig. 1. In this paper,

we only describe how to determine the key variables for road-

departure test scenarios. The determination of the most

representative values for key parameters will be presented in

a separate paper.

Key parameters for road departure test scenarios can be

obtained from the distribution of road departure conditions

associated with run-off road crashes. These conditions contain

variables including vehicle speed, road edge type, vehicle

departure angle, and environmental factors. It is desirable that

these conditions can be used to generate key parameters for

RDDS testing. In this work, possible data sources containing

conditions of road departure crashes were identified and

analyzed. These data sources are from National Automotive

Sampling System Crashworthiness Data System (NASS

CDS), Fatality Analysis Reporting System (FARS), National

Motor Vehicle Crash Causation Survey (NMVCCS), and so

on.

In these databases, many factors/conditions are associated

with the descriptions of road departure crashes. These factors

are: (1) road conditions including road geometries, road

surface, road slope, and radius of the curvature; (2) roadside

conditions including road edges and boundaries; (3) vehicle

conditions including vehicle departure speed, vehicle road

departure angle, vehicle lateral speed, and side of road that the

departure occurred; (4) environment conditions including

weather, time, and lighting conditions; and (5) driver

attentiveness such as distraction and fatigue. Since we are

interested in the effectiveness of road-departure

detection/mitigation systems, factor (5) is not applicable.

Therefore, seven key parameters for describing general road-

departure test scenarios are selected as follows:

(1) Road type – the road geometries or alignment (straight

road or curved road);

(2) Radius of road curvature – radius of curved road;

(3) Road edges type – roadside boundaries and materials,

such as grass, gravel, concrete curb, metal guardrail, and

so on;

(4) Vehicle departure velocity;

(5) Vehicle departure angle – the relationship between

vehicle departure speed and vehicle departure angle are

depicted in Fig. 2, where vehicle CG (center of gravity)

velocity is the same as vehicle departure speed;

(6) Vehicle departure side – on which side of the road that

road departure occurred; and

(7) Lighting conditions – time, weather condition, and street

lighting.

III. STRUCTURE OF ROAD DEPARTURE TEST SYSTEM

For the development of standardized performance

evaluation of road-departure detection systems, the

coordinated test data collection and post-processing systems

were designed and implemented. Fig. 3 depicts the proposed

overall structure for road departure detection system testing.

The system includes: (1) a test vehicle with a data collection

system, (2) surrogate roadside objects, including grass, metal

guardrail, curb, and concrete divider, and, (3) a central

computer for data recording and system coordination and

Fig. 1. Overall approach for road departure test scenarios.

Fig. 2. Relationship between vehicle departure speed, lateral

speed, and departure angle.

Fig. 3. Proposed structure of road departure testing system.

1374