CLIENT
For more than 90 years, COWI has been one of Denmark’s leading consulting companies specializing in engineering, environment, and economics. They have offices all over the world, around 7,200 employees, and are always working on 12,000+ projects.
A special focus area is professional consultation in connection with e.g., construction projects, study of traffic conditions, etc., including providing the necessary analysis and documentation of existing conditions, development, and results.
It was on the basis of this and as a further development of their service that COWI launched DDG in 2016. DDG is an “on demand” service for professional use with HD panoramas of the Danish road network developed from image material collected by COWI’s cars.
Ensuring accuracy, the maintenance of DDG entails a complete photographic documentation of the entire Danish road system every two years. Every street, road, avenue, and boulevard – all 108,000 km of road network – is documented. In total, this amounts to almost 150 single or 30 million panoramic images in HD quality.
CHALLENGE
GDPR
In relation to the GDPR, the specific task was that neither persons nor number plates on vehicles must be identifiable. It must not be recognizable who or what vehicle has been captured by the photographer. Both faces and license plates must therefore be blurred.
This represented two sub-tasks: firstly, making it possible to determine whether it was a face or a number plate at all, and secondly, launching the blurring process efficiently, quickly and with an acceptable result.
Up higher – Price lower
COWI had an overall ambition to upgrade the existing DDG set-up with the purpose of getting a more flexible, scalable, and cost-effective solution, where the workflow for processing the image material was automated to a much greater degree.
SOLUTION
KeyCore was chosen as a partner for the design of a cloud infrastructure, in which the processing of image material could take place more cost-effectively and scalable compared to the existing on-prem solution.
To handle this, the processing of the image material for DDG was migrated to an AWS cloud infrastructure.
The setup of COWI’s cloud infrastructure was orchestrated in accordance with the strict guidelines of GDPR, so that storage falls into two stages. According to GDPR, it is not permitted to store non-anonymised footage of this nature for more than six months. Therefore, it was crucial that images could automatically go through the anonymization process immediately after upload.
With the AWS S3 Standard, COWI has gained access to maximum capacity during the short period image material is being processed. At the same time, S3 has automated backup, which gives a guarantee of 99.999999999% of the material on an annual basis, or a maximum loss of 1 object out of 10,000,000 in 10,000 years.
The long-term storage is handled in an AWS S3 Glacier – which only costs $ 0.004 per. gigabyte pr. month – and AWS Batch, which ensures both scalability and access to resources and speed, yet maintaining never seizing more resources than necessary and using them when the price is lowest.
Security
By migrating to the AWS cloud, COWI has gained access to the highest possible level of compliance for everything below the hypervisor line. This includes infrastructure, hardware, the physical facilities, and networks. It is audited and documented by third parties several times a year and users of AWS can include it in their documentation proving that they are compliant and obtain the necessary certification that may apply to them. This is compliance with, among others, ISO 27001, GDPR, EU-US Privacy Shield, HIPAA, and HMDA.
Post processing
Part of the post processing of the image material had previously been done manually, which not only tied up a large amount of resources, but also led to an increased risk of errors.
With the cloud infrastructure orchestrated by KeyCore, COWI was able to ensure that the entire process was automated and streamlined using ML.
Specifically, the KeyCore-architecture gave COWI access to use the market’s absolute leading algorithms in a highly modular container-based setup, where COWI have easy access to replace modules in accordance with the development of the models.
The use of ML includes several different Deep Learning Convolution Neural Networks (CNN), where the neural networks are “trained” on already classified images and subsequently can classify and identify new material.
With the cloud infrastructure developed by KeyCore, the post processing has become much more automated, which ensures a fast and efficient process from when the image material is uploaded until it is ready for publication in DDG.
In addition to ensure the GDPR-demanded blurring of faces and license plates, the post processing in the cloud also includes stitching of the panoramic images, correction of the image quality and a dashboard for monitoring the status of the processing of uploaded material, which gives COWI’s clients a valuable tool in their professional work.