Artificial Intelligence for an efficient data anonymisation solution

Client background and business context

Client: European leader that offers comprehensive collections of infrastructure data, whether classic or through navigation or aerial survey

Client’s data: the only provider on the market that successfully offers, uses, and combines all data acquisition methods, with over 40 years of experience

Project: object detection with AI, automatic number plate and face detection solution, including the anonymisation of private data

Challenges:

With a strong focus on municipalities, utilities, construction companies as well as companies from the industrial and tourism areas, our client must collect and anonymise large collections of data, be that private or not private.

Being an engineering and surveying office, but also a technology-leader for mobile road data acquisition, our client ensures the end-to-end process.

Thus, for this project, our main challenges were in the areas of:

Anonymising images for data protection: create a tool capable to automatically anonymise large data collections for enhanced privacy in a timely manner

Automating the processing of gathered images: switch from the human-made anonymisation process to a digital one, enabling the human factor to have more of a decisive post-processing role

Increasing the company’s delivery speed: ensure a 50x faster data processing and anonymisation time with a velocity of 0.3 sec/image

Ensuring a higher quality of the image processing endeavour: decrease the number of processing errors and ensuring a high quality of the end-results

Future-readiness: enable the digital solution to grow and develop, as the company’s needs diversify

Our approach
For our projects, we developed a five-step approach

In the first phase of Discovery, our specialists made a thorough competitional analysis of similar products existing on the market. After the analysis, we have discovered that none of the solutions could respond to our customer’s needs in terms of processing time and quality.

The approach was first validated through a prototype. For this, we focused on meeting the success criteria defined by our clients in terms of accuracy and speed. With the prototype defined and validated by the stakeholders, we were ready to move the project into production.

In the production phase, the model was integrated into the already existing platform of our client. The processes on the customer’s end were validated through the monitoring of the data flow and outputs.

In terms of UI, we have developed a user interface that further facilitates human validation of results, as well as other interventions if and when needed.

Relevant services:

Technologies used:

Python

TensorFlow

Keras

OpenCV

Results

After the implementation, the algorithm ensured a 90% reduction of the overall manual processing.

In this context, the human factor switched from the doer factor to that of a reviewer.

The previous human-made anonymisation process took an average of 15s per picture. After the implementation, this processing time decreased to 0.3s per picture. This is a 50x reduction and time optimisation with impact in both the delivery time and in the decrease of overall processing costs.

The client expected an overall accuracy of 80% and a processing time of 1s. For this project, we have surpassed the client’s expectations in terms of accuracy and processing time, as we managed to ensure a 92% accuracy with a 0.3s processing time.

We enabled a future-ready solution that allows the development and deployment of new integration whenever needed. As the project evolves and grows, extra features will be added for increased efficiency, data quality and to cover a larger spectrum of needs.

Success stories:

Asset management with IoT for enhanced efficiency, compliance, and cost-optimisation
Asset management
with IoT
Manufacturing execution systems​ customisation
MES
Customisation