Computer aided tools for automation of industrial inspections**

*基础设施是一家在魁北克运营的工程咨询公司。该公司拥有160多名员工，在公共事业基础设施和外部电信网络的工程和设计方面处于领先地位。在本项目中，我们提出了三种不同的系统来支持电信土建工程的测量。第一个是我们对进入井的工作的继续。目前的样机存在图像质量不高、视场有限等局限性。我们建议将常规高清晰度相机集成到该系统中，并通过开发新的稳健测量算法来提高其精度。同时，我们的目标是建立一个关于进入井的信息的数据库。*第二个系统侧重于管道安装的地形检查。在施工期间跟踪地形轮廓及其不规则性以及新管道和现有管道的位置是一项基本任务，因为未强制实施的地形可能会因雨或风而每天发生变化。我们提出了一种系统，可以在这样的地形上进行测量，并在现场检查信号差异期间自动进行测量，并帮助测量工程师准备报告。*第三个系统侧重于电线杆。这是一项非常具有挑战性的任务。我们提出了一种深度学习解决方案，其中神经网络被训练来学习正确的电极点是什么样子，并且它基于统计分布来检测异常。深度神经网络实时或作为后处理步骤将检测可能具有不正确配置的对象，并将它们通知操作员。对于无法进入的区域，我们将考虑使用无人机进行测量。将自动创建一个包含现有极点信息的数据库。****

**Infrastructel is an engineering consulting firm operating in Quebec. It employs over 160 people and is a leader in engineering and design of public utility infrastructure and outside telecommunications networks. In this project, we propose three different system to support the surveying of telecommunication civil works. The first is a continuation of our work with access wells. The current prototype has some limitations such as low-quality imagery and limited field of view. We propose to integrate to this system with a regular high-definition camera and increase its accuracy by developing new robust measuring algorithms. In the same time, we aim to build a database with the information regarding the access wells.****The second system focuses on terrain inspection for ducts installation. Keeping track of a terrain profile and its irregularities and the position of the new and existing ducts during construction is an essential task as the unenforced terrain can shift from day to day due to rain or wind. We propose a system that can take measurements on such a terrain and automatically during field inspection signal differences and assist the surveying engineer in preparing the report. ****The third system focuses on utility poles. This is a very challenging task. We propose a deep learning solution where a neural network is trained to learn what a correct utility pole looks like and it detects anomalies based on a statistical distribution. A deep neural network, either in real-time or as a postprocessing step will detect the objects that are likely to have incorrect configurations and will signal them to an operator. For inaccessible areas, we will consider using drones for surveying. A database with the information from the existing poles will be created automatically. ****