Description of the privacy protection

Description of the object recognition process

The AI model performs real-time object detection without saving images.

The AI model utilises a series of real-time object detection systems based on convolutional neural networks. The algorithm requires only one forward propagation pass through the neural network to make predictions, unlike region proposal-based techniques that require thousands for a single image.

Edge computation: to ensure complete privacy protection, the AI model is run as an edge computation on a mobile phone, where no personal data is stored or transmitted.

Real-Time Inference Without Storage: The AI model performs object detection by processing each video frame on the fly, identifying and counting objects (people) within milliseconds.

Object recognition process without personal data: the AI model has been trained with training material including only photographs of annotated shapes of people, which are fully anonymised. Faces are not recognisable on the training material and due to the nature of having photographs from the point of view of top-down angle, the training data does not include any recognizable physical characteristics  such as height, body shape, or distinctive marks (e.g., tattoos, scars).

It has not been including or annotated any clothing or accessories (e.g., a uniform with a name tag) that can identify an individua). Further the training material does not include any meaningful contextual information such as identified places. The training data photographs backgrounds has been in a significant manner made of synthetic data to ensure that no recognisable features of background are trained. The AI model therefore has been trained only to recognise generic shape of people and therefore no personal data is processed on the execution of our AI model.

Data Output: After inference, the model stores only following metadata: number of objects, direction and time. In this way, the system collects information about the number of people without storing any image data. The output of the AI model contains only statistical data. The solution does not process or store any personal data and should not be subject to the General Data Protection Regulation (GDPR).

The system does not to fall under the scope of GDPR for the following reasons:

Customer's responsibility to maintain the anonymity of the data

It is the responsibility of the customer not to combine data received from us with other data sources and thereby create personal data without an appropriate risk assessment as required by the GDRP. While such data creation is external to our system, we will nevertheless always actively seek to inform customers of any risks we identify in the processing of their data and, where appropriate, recommend that we transform the data we provide to them into a more statistical format to minimise any risks external to our system that may be present in the customer's data management.