Automated Anonymization of Visuals with Computer Vision
Vaisala’s anonymization service is capable of detecting and anonymizing vehicles and pedestrians in images and video material. This technology opens up interesting opportunities for many, as it allows for more efficient use of visual data and the ability to fully leverage its benefits.
Today, organizations are met with the challenge of ensuring data privacy and complying with various regulations such as the General Data Protection Regulation (GDPR) recently implemented and applicable to all the member of EU. Vaisala’s own computer vision engine uses the latest technology and models to deliver the greatest value possible. The automated anonymization of visual material is achieved through semantic segmentation. Anonymization accuracy is constantly pushed as Vaisala progressively develops the models for object detection. At the moment the various test material and customer deliveries demonstrate close to 99% accuracy in videos recorded in public spaces.
Anonymize Only What Needs to be Anonymized
It is crucial that anonymization provides clear images, leave other valuable data objects untouched, and does not interfere with other computer vision detectable objects; such as traffic signs, cracking or potholes, as the processed visual material is used for simultaneous monitoring of multiple infrastructural elements. An important objective has also been to achieve the ability to anonymize material at a large-scale in order to deliver on customer needs of any proportion. This was realized through an actual process through which computer vision detects the objects, and rapid assessment of bulk imagery is accomplished by scaling the annotated images down.
For the required anonymization, computer vision had to be taught to reliably identify and label people or vehicles. This required annotation of extensive data sets with labels applied to objects in images captured at a street-level. Semantic segmentation of these images was then used to develop and instill object detection of people and vehicles. The computer vision model's accuracy continues to improve as new dataset images are collected and annotated.
Success with Departments of Transportation
Anonymization has proven itself priceless for Departments of Transportation (DOT) held accountable for compliance with the regulations regarding data protection and privacy. Previously, visual material of DOT networks included public users, which resulted in data that could not be shared with, for example, partners responsible for certain operations or segments of the network. With automated, computer vision anonymization, time and cost-efficient results are achieved in near real-time. This allows DOTs and their partners to continue their operations with the support of visual material, yet without having to worry about potential data privacy breaches.
Contact us to get more information on how anonymization can benefit you.