VideoPipe 2022 Challenge Real-World Video Understanding for Urban Pipe Inspection Yi Liu1 Xuan Zhang1 Ying Li1 Guixin Liang2 Yabing Jiang2 Lixia Qiu2 Haiping Tang2

2025-05-06 0 0 4.16MB 7 页 10玖币

侵权投诉

VideoPipe 2022 Challenge: Real-World Video

Understanding for Urban Pipe Inspection

Yi Liu1∗, Xuan Zhang1∗, Ying Li1, Guixin Liang2, Yabing Jiang2, Lixia Qiu2, Haiping Tang2,

Fei Xie2, Wei Yao3, Yi Dai2†, Yu Qiao1,4†, Yali Wang1,5†

1ShenZhen Key Lab of Computer Vision and Pattern Recognition, Shenzhen Institute of Advanced Technology,

Chinese Academy of Sciences, China

2Shenzhen Bwell Technology Co., Ltd, China

3Shenzhen Longhua Drainage Co., Ltd, China

4Shanghai AI Laboratory, Shanghai, China

5SIAT Branch, Shenzhen Institute of Artiﬁcial Intelligence and Robotics for Society

Abstract—Video understanding is an important problem in

computer vision. Currently, the well-studied task in this research

is human action recognition, where the clips are manually

trimmed from the long videos, and a single class of human action

is assumed for each clip. However, we may face more complicated

scenarios in the industrial applications. For example, in the real-

world urban pipe system, anomaly defects are ﬁne-grained, multi-

labeled, domain-relevant. To recognize them correctly, we need

to understand the detailed video content. For this reason, we

propose to advance research areas of video understanding, with

a shift from traditional action recognition to industrial anomaly

analysis. In particular, we introduce two high-quality video

benchmarks, namely QV-Pipe and CCTV-Pipe, for anomaly in-

spection in the real-world urban pipe systems. Based on these new

datasets, we will host two competitions including (1) Video Defect

Classiﬁcation on QV-Pipe and (2) Temporal Defect Localization

on CCTV-Pipe. In this report, we describe the details of these

benchmarks, the problem deﬁnitions of competition tracks, the

evaluation metric, and the result summary. We expect that, this

competition would bring new opportunities and challenges for

video understanding in smart city and beyond. The details of our

VideoPipe challenge can be found in https://videopipe.github.io.

I. INTRODUCTION

In the last decades, sewer pipe system is one of the most

crucial infrastructures in modern cities. In order to ensure

its normal operation, we need to inspect pipe defects in an

effective and efﬁcient manner. Several technologies have been

applied in the traditional pipe inspection procedure. [1] has

conducted a thorough investigation and categorized them into

visual methods, electromagnetic methods, acoustic methods,

and ultrasound methods. In particular, Quick-View (QV) In-

spection and Closed-Circuit Television (CCTV) Inspection are

the most popular methods, as shown in Figure 1. The Quick-

View (QV) Inspection is used for rapid anomaly assessment

on sewer pipes, since the camera can only record videos

on the pipe oriﬁce. The CCTV inspection system involves

a remote-controlled robot that travels along the sewer pipe

with a camera for video recording [2]. Hence, it can get more

∗Yi Liu (yi.liu1@siat.ac.cn) and Xuan Zhang (xuan.zhang1@siat.ac.cn)

are equally-contributed ﬁrst authors.

†Yi Dai (daiyi@bominwell.com), Yu Qiao (yu.qiao@siat.ac.cn) and Yali

Wang (yl.wang@siat.ac.cn) are equally-contributed corresponding authors.

Fig. 1: Two widely-used pipe inspection methods.

detailed anomaly analysis for the whole pipe. Based on these

QV and CCTV videos, the standardized protocols for manual

inspection have been established and adopted in the recent

years [3]. However, it is often labor-intensive to ﬁnd anomaly

from hundreds of hours of videos in the complex urban pipes.

To tackle this problem, it is essential to develop automatic

inspection methods to discover sewer anomaly from large-

scale pipe videos. Early works use hand-crafted visual features

with traditional classiﬁers [4], [5], [6]. These approaches are

often limited for inspecting defects in the complex scenarios.

arXiv:2210.11158v1 [cs.CV] 20 Oct 2022

Fig. 2: Anomaly Examples in Our QV-Pipe Dataset.

Fig. 3: Data Distribution of QV-Pipe Dataset.

With fast development of deep learning frameworks, a number

of neural networks have been recently applied for defect classi-

ﬁcation [7], [8], [9], [10], [11], [4], [12], defect segmentation

[13], [14], [15], and sewer defect detection [16], [17], [18],

[19], [20], [21], [22]. However, these methods are mainly

based on image-level detection. Hence, they are difﬁcult to

ﬁnd multi-labeled and ﬁne-grained defects, without learning

spatial-temporal contexts in the video. Moreover, these data

sets are not available for academic research, which blocks to

develop advanced learning approaches for reliably inspecting

sewer defects.

Based on the discussions above, we propose to establish

real-world video understanding for urban pipe inspection. To

achieve this goal, we carefully collect and annotate two new

industrial video datasets, namely QV-Pipe and CCTV-Pipe,

for our VideoPipe challenge. Speciﬁcally, QV-Pipe is used for

video defect classiﬁcation (Task 1) and CCTV-Pipe is used for

temporal defect localization (Task 2). In this paper, we will

introduce the details of data and tasks, and summarize the

competition results. We expect that, this competition would

bring new opportunities and challenges for video understand-

ing in smart city and beyond.

II. DATASETS AND ANNOTATION

In this section, we present how these two datasets are

collected and how they are constructed. Note that, these

datasets are based on the real-world pipe networks. Hence,

we have deleted the information of street, city and any other

about privacy in our datasets.

A. QV-Pipe Dataset

The QV-Pipe dataset consists of 9.6k videos, which are

collected from real-world urban pipes. The total duration of all

videos exceeds 55 hours. Moreover, there are 1 normal class

and 16 defect classes. Because the pipe situation is complex

and multiple defects often appear at the same time, each video

is annotated with multiple labels. The professional engineers

are required to do this annotation procedure. To obtain accurate

annotations of defect instances, these engineers are asked to

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：

VideoPipe2022Challenge:Real-WorldVideoUnderstandingforUrbanPipeInspectionYiLiu1,XuanZhang1,YingLi1,GuixinLiang2,YabingJiang2,LixiaQiu2,HaipingTang2,FeiXie2,WeiYao3,YiDai2y,YuQiao1;4y,YaliWang1;5y1ShenZhenKeyLabofComputerVisionandPatternRecognition,ShenzhenInstituteofAdvancedTechnology,ChineseAcade...

展开>> 收起<<

VideoPipe 2022 Challenge Real-World Video Understanding for Urban Pipe Inspection Yi Liu1 Xuan Zhang1 Ying Li1 Guixin Liang2 Yabing Jiang2 Lixia Qiu2 Haiping Tang2.pdf

共7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！

VideoPipe 2022 Challenge Real-World Video Understanding for Urban Pipe Inspection Yi Liu1 Xuan Zhang1 Ying Li1 Guixin Liang2 Yabing Jiang2 Lixia Qiu2 Haiping Tang2

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: