1 Paper ID Vision -based Warning System for Maintenance Personnel on Shor t-Term Roadwork Site

2025-04-28 0 0 1.14MB 8 页 10玖币

侵权投诉

Paper ID #

Vision-based Warning System for Maintenance Personnel on Short-Term

Roadwork Site

Xiao Ni1*, Walpola Layantha Perera1, Carsten Kühnel1, Christian Vollrath1

1. University of Applied Sciences, Altonaer Straße 25, 99085 Erfurt, Germany

*xiao.ni@fh-erfurt.de

Abstract: We propose a vision-based warning system for the maintenance personnel working on

short-term construction sites. Traditional solutions use passive protection, like setting up traffic cones,

safety beacons, or even nothing. However, such methods cannot function as physical safety barriers to

separate working areas from used lanes. In contrast, our system provides active protection, leveraging

acoustic and visual warning signals to help road workers be cautious of approaching vehicles before

they pass the working area. To decrease too many warnings to relieve a disturbance of road workers,

we implemented our traffic flow check algorithm, by which about 80% of the useless notices can be

filtered. We conduct the evaluations in laboratory conditions and the real world, proving our system’s

applicability and reliability.

Keywords: Traffic Safety, Computer Vision

1. Introduction

The safety of road maintenance workers has been a well-known object of traffic research and

development projects in Europe in recent years. Guidelines for human behavior, roadwork site setups,

and technical solutions were developed and implemented [1, 2, 3]. Almost all of these guidelines deal

with preventing rear-end collisions with safety trailers. Another issue that has not been dealt with yet is

the safety of maintenance personnel, especially their safety within the area of the short-term roadwork

sites (STRWS). When the maintenance workers are working, sometimes they cannot sense the

surroundings explicitly. Moreover, the construction vehicles in the front can obscure the view of the

road workers, and they cannot see the traffic behind them.

Figure 1: Short-term roadwork site

In many European countries, the short-term roadwork sites are separated from lanes with free-flowing

traffic only by mobile warning signs such as traffic cones, safety beacons, or even nothing (Figure 1).

Protection by passive protective devices is not possible due to the limited duration, and no physical

barriers can hold back vehicles in case of an accident. As a result, personnel in STRWS are generally

exposed to higher risks. Therefore, the German Federal Ministry of Labour and Social Affairs

implemented a guideline to lower the risk of accidents in short-term roadwork sites. However, Hands-

on experience shows that this is difficult to observe for maintenance workers, as there are many

situations in which the “No-Entry Zone” is entered, consciously or unconsciously.

Our solution addresses this problem by trying to detect the approaching vehicle from both front and rear

directions based on computer vision and trigger a warning signal to alert the maintenance personnel. To

reach that aim, we utilize the recent groundbreaking deep learning-based Object detection for detecting

approaching vehicles on the road in both directions. Furthermore, the system can track such vehicles

and record their trajectories. After tracking vehicles, we utilize our traffic flow check algorithm to filter

unnecessary warnings.

Figure 2: Architecture of the vision-based warning system

There are two inputs to the system. One is a video streaming of a front-mounted conventional RGB

camera, and another is a video streaming of the rear-mounted conventional RGB camera. With these

two cameras, we can observe both directions of the construction site. The whole system is composed

of three main components: (1) a fast detector, (2) a detector-based tracker (3) a traffic flow check

algorithm (see Figure 2).

2. Related works

Vehicle detection is a considerably mature and proven technology that is crucial for the whole system.

Traditional detectors combine the sliding window method with traditional machine learning classifiers

[4, 5]. Such methods rely on handcrafted features, require much feature engineering work, and cannot

achieve satisfactory accuracy. In recent decades, due to the groundbreaking improvements in deep

learning, modern detectors are mainly composed of the convolutional neural network (CNN) [6, 7, 8].

CNN has a similar property to the traditional fully connected network but considerably reduces the

amount of the neural network’s parameter, making CNN much more robust and less prone to overfitting.

Such CNN-based detectors require large-scale datasets to train their neural network and can achieve a

better generalization capability. As a result, their accuracy is significantly improved in comparison to

traditional detectors. Since the working condition of maintenance is outdoor and constantly varies from

one country road to another, some country roads’ backgrounds are dense forest and mountain bodies,

while the others ought to be open fields. The weather can also vary seasonally, from rain to sunshine.

Therefore, we need a robust and highly reliable detector for the application and choose a CNN-based

detector. CNNs have a high requirement of computing resources to get a fast inference speed, so in the

project, we equipped the vehicle computer with GPU, which can accelerate the neural network’s

inference.

There are also two main branches of modern detectors. The first is R-CNNs based on the regional

proposal, and the inference process consists of two stages [6, 9, 10]. The other is a one-shot method,

which operates the feature extraction and object localization at the same time [7, 8, 11]. The one-shot

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

1PaperID#Vision-basedWarningSystemforMaintenancePersonnelonShort-TermRoadworkSiteXiaoNi1*,WalpolaLayanthaPerera1,CarstenKühnel1,ChristianVollrath11.UniversityofAppliedSciences,AltonaerStraße25,99085Erfurt,Germany*xiao.ni@fh-erfurt.deAbstract:Weproposeavision-basedwarningsystemforthemaintenanceperson...

展开>> 收起<<

1 Paper ID Vision -based Warning System for Maintenance Personnel on Shor t-Term Roadwork Site.pdf

共8页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

1 Paper ID Vision -based Warning System for Maintenance Personnel on Shor t-Term Roadwork Site

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: