Renfe wants to monitor travelers with facial recognition in real time: what this system will be like and why it can be discriminatory
Renfe will use facial recognition to control the capacity, detect who sneaks and make a profile of travelers. It’s about a pilot project which intends to use the 200 cameras installed in the main stations throughout Spain. At the moment these images are only used by security agents to control, but Renfe wants to implement an analysis of the images in real time using algorithms based on artificial intelligence. That is, to apply an automated facial recognition system that allows them to receive alerts if a user passes under the doors or to have control of what type of people are at each station.
The national train company wants to have the system implemented by May 2021 and therefore has opened a tender worth 781,000 euros to develop the image analysis system that will be implemented in 25 Cercanías stops. A proposed system that, as it has advanced The confidential, will allow classify travelers by age, gender, mood, and even ethnicity. The latter is a particularly sensitive data protected by data protection laws.
Up to 98% effective facial recognition
As described in the bidding document, which Xataka has been able to access, the system must be able to analyze in real time the capacity of the most relevant Cercanías stations in the centers of Madrid and Barcelona. In total, the system is designed to be used in 25 stations, mainly those of Madrid and Barcelona capital, but also those of Getafe, Fuenlabrada, Alcorcón, Terrasa, Mataró, San Bernardo, San Isidro and Llodio.
This facial recognition system is created to automatically trigger alarms to the C24h control centers of the Department of Safety, Self-Protection and Prevention of Occupational Risks. The algorithm will analyze the images captured by the cameras of these stations and will allow detecting certain actions, with a minimum efficiency required by Renfe.
The Renfe project differentiates between the Andenes area and Los Tornos. For platforms, the system will perform a counting of travelers, detecting maximum capacity every 30 seconds. In the turnstile area, the facial recognition system must be able to count the entrances and exits of each turnstile, detect fraud, perform an analysis of the type of luggage (small / large / medium suitcase) and create a traveler profile (age, gender, ethnic origin).
To control capacity, Renfe requires a minimum efficiency of 98%. However, for the detection of fraud, such as jumping, making the train, going under or between the gates, Renfe requests 95%. A percentage that drops to 85% in relation to making the profile of travelers. If, during the execution of the contract, Renfe finds that the degree of efficiency required for the analytics is not met, the successful bidder could be penalized and even be cause for termination of the contract.
The segmentation that this facial recognition system wants to do is especially striking, where it will not only be differentiated by age or gender, but also by ethnic origin, type of clothing, whether to travel alone, with children or in large groups and even the mood (happy, sad, tired, undetermined).
The analysis is not there, since after the first six months of the contract, it is expected that antisocial behaviors such as riots, fights and disturbances, as well as anomalous behaviors, can also be analyzed. vandalism, graffiti, loitering or theft. Once these data have been obtained, the document refers to the fact that an analytical system will be implemented that allows the “advance prediction of negative events and repetitive behaviors with the study of patterns “.
What Data Protection says about distinguishing by ethnic origin
Samuel Parra, a data protection specialist at Égida, explains that ethnic origin is especially protected by the 2018 Data Protection Regulation, as it is a special category of data. In this aspect, the consent of Renfe users would not be enough for facial recognition to identify them by origin. That is, facial recognition could identify gender or even mood, we assume that through facial expressions, but it would not be acceptable for the algorithm to take into account racial origin, in addition to difficulties involved.
According to article 9 of Special categories of data, of the RGPD:
In order to avoid discriminatory situations, the consent of the affected person will not be enough to lift the prohibition of data processing whose main purpose is identify your ideology, union affiliation, religion, sexual orientation, beliefs or racial or ethnic origin.
This article has some exceptions, such as an “essential public interest”. Something that would not be fulfilled in this case, according to Parra. From Xataka we have consulted with Renfe to find out their position in this regard, with no response at the moment.
Renfe cancels the initial tender to make corrections
Within a few hours of announcing the tender through El Confidencial, Renfe has canceled the specifications of its facial recognition system, in order to “make corrections in the Specifications and in the Technical Specifications document”. The document is no longer accessible, although it will be expected to be updated in the coming weeks when it has been revised.
Beyond the ethnic origin aspect, the truth is that Renfe does take Data Protection into account in its tender. It is striking that in its Dashboard, it is explained that data such as season, year, month, week, day, gender and age will be filtered by real time, but neither mood nor ethnic origin is described. This last distinction could easily be abandoned by Renfe and the vast majority of the 170-page document would remain valid, according to experts.
Among the requirements requested by Renfe the creation of a Data Protection Officer is specified, as well as the performance of a risk analysis which will be shared with the Renfe Group’s Cybersecurity and Privacy Area. In section PS.07, it is specified that the Organic Law on Data Protection and Digital Rights Guarantee must be complied with, something that according to the experts conflicts with the requests made by Renfe itself in this tender.