Improved noise pollution forecasting using machine learning methods
Using camera technology and machine learning, the engineering firm Richters & Hüls can create noise emission forecasts based on robust data. Employees are freed from the burden of time-consuming data collection and can use their expertise to add value.
The Richters & Hüls Engineering Office has been operating in the field of technical environmental protection, especially in the fields of waste management and pollution control, as well as approval procedures subject to the Federal Immissions Control Act (BImSchG) since 1991. Focus topics include the preparation of odor, ammonia, dust, and noise forecasts. The forecasts provide the basis for approval procedures for construction projects for customers from agriculture, industry, commerce, and the public sector.
The project at a glance
- Continuous, automated data collection instead of manual recording of visitor frequency by counting
- Reduced data collection costs and improved data quality
- Data protection-compliant counting
- Easy access to analysis and reporting counting data via API
Initial situation
Immission control reports are prepared as part of new construction projects or the redesign of existing facilities to ensure that no harmful environmental impacts will occur. When a new soccer field is laid out, for example, emission data is determined and used as a baseline for an immissions forecast. Visitor frequency, number of visitors, times of visits and length of visits are important and essential for determining emissions. However, current manual methods are time-consuming and prone to statistical inaccuracies.
The Richters & Hüls Engineering Office 
was commissioned to produce an immissions control report for a football pitch in a residential area. The football pitch is to be moved because of a new building in the area. The engineering firm wondered how the data necessary for the emission forecasts could be determined automatically.
Immission forecasts are based on assumptions or statistical surveys. Counting visitors, for example on a playground or football pitch, is very complex and usually is not conducted over an extended period. Therefore, instead of an exact count, visitor frequency is determined by counting at selected times.
This statistical approximation method is now to be replaced by an automated system in order to obtain actual visitor numbers. With a better data base, stakeholders can make better decisions about construction projects.
Solution
Setup
On-site Installation
Data protection compliant thanks to edge installation
Setup
In the project, codecentric was responsible for selecting and setting up the hardware (camera, controller). The codecentric project team also developed the software for the automated counting of people using machine learning and set up a cloud environment where assessors can retrieve the count data exclusively. Our partner, Phoenix Contact, a leading component and system manufacturer, supplied the appropriate hardware.
The PLCnext AXC F 3152 controller from Phoenix Contact is used on site. The controller runs a Linux operating system, is IP20 certified and compatible with a new extension to accelerate machine learning applications. This extension has an edge TPU, making it ideal for machine learning applications directly on the device.
Any network camera that communicates via a common protocol can be used as an optical sensor. The controller analyzes the camera image directly. A machine learning model classifies all objects in the camera image. All objects with the class label "human" are counted. This counter value plus timestamp is then transferred to the cloud via MQTT – an open network protocol for machine-to-machine communication – where it is available to be accessed for analysis and reporting. The use of a pre-trained machine learning model meant that no time needed to be invested in training or collecting training data.
On-site Installation
Only two appointments were necessary to set up the system. The first on-site visit was to assess the feasibility and select a location for the optical sensor used. The sensor was installed and put into operation on the second visit.
Finding a suitable location for the optical sensor that would allow it to be used in the long term was not easy. The location that was originally considered – behind a window in a building near the football pitch – turned out to be less than ideal. The glass pane caused reflections and opening the window created unwanted camera movements. The solution was to install the sensor on the roof of a building directly adjacent to the pitch, from where there was an unobstructed view of the football pitch, the camera was not in anyone's way, and it was possible to install it permanently right next to the power supply.
Data protection compliant thanks to edge installation
The entire solution is data protection-compliant. All the information collected, in particular the camera image, is not stored in order comply with data protection requirements. No image is sent to the cloud for analysis or to any other powerful computing instance. Everything is performed “on the edge”, i.e. on the controller installed on site. Only the evaluated count data is stored for analysis and reporting.
On-site Installation
Only two appointments were necessary to set up the system. The first on-site visit was to assess the feasibility and select a location for the optical sensor used. The sensor was installed and put into operation on the second visit.
Finding a suitable location for the optical sensor that would allow it to be used in the long term was not easy. The location that was originally considered – behind a window in a building near the football pitch – turned out to be less than ideal. The glass pane caused reflections and opening the window created unwanted camera movements. The solution was to install the sensor on the roof of a building directly adjacent to the pitch, from where there was an unobstructed view of the football pitch, the camera was not in anyone's way, and it was possible to install it permanently right next to the power supply.
Result
The engineering firm now benefits from more affordable options for determining emissions data through the automated and cost-effective collection of visitor frequency.
Data-driven analysis allows for more precise work. Decisions on planned sports and leisure facilities are based on a valid and solid data base.
The experts at the Richters & Hüls Engineering Office will in future be able to apply their expertise in environmental protection and technology in an even more targeted way. They save time in gathering the necessary data and can therefore focus more on their analysis.
Dr. Jannik Hüls
Managing Director, Engineering Office Richter & Hüls
For us, the application is an example of how technology can support people. Thanks to the automated solution, we were able to produce the report faster, more cheaply, and also more precisely for our customer, the German Red Cross in Coesfeld.
Dr. Jannik Hüls
Managing Director, Engineering Office Richter & Hüls
Do you have more questions about the project?
Would you like to learn more about the project? Let's meet and talk.
Management Board Consulting / Head of Sales
Other projects
Find out about other successful projects that we have completed with our clients. Perhaps you will find ideas here for a use case in your own organization.
Markus Zink
Management Board Consulting / Head of Sales