PF uses artificial intelligence to identify vandals – 01/24/2023 – Politics

The Federal Police use a mixture of artificial intelligence and manual work by police officers to identify supporters of former President Jair Bolsonaro (PL) involved in the attacks on the Congress buildings, the Planalto Palace and the STF (Federal Supreme Court).

The material produced by experts from the INC (National Institute of Criminalistics) and papilloscopists from the INI (National Institute of Identification) supplies the operations carried out after the afternoon of vandalism on January 8th and will be used in the investigations opened to investigate the attack against the three Powers.

According to reports by investigators, the papilloscopists are acting more in the short term, to arrest suspects who were not detained on the 8th or the day after the closure of the camp in front of the Army headquarters.

In these cases, they use identification software and perform the so-called prosopographic examination, in which a comparison of the elements of the face is made to assign compatibility or incompatibility with those presented in images or photos.

Papilloscopists acted, for example, in identifying some fugitives, such as the man who broke Dom João’s watch in the Planalto Palace. The suspect, identified as Antonio Claudio Alves Ferreira, 30, was arrested this Monday (23) in Uberlândia (MG).

The PF has already ruled out, through this type of analysis, the possibility of him being a member of the MST (Movement of Landless Rural Workers) as it had been disseminated on the networks.

While the INI works on these specific cases, a team of around ten experts from the INC’s audiovisual sector uses artificial intelligence developed by professionals from the PF itself to produce a report with all the dynamics of the attacks and to identify, one by one, the scammers that appear in the closed-circuit cameras in the buildings, videos collected on cell phones of prisoners and images collected in the investigations.

Only from the Planalto Palace, according to the Sheet, images were collected from 22 cameras of the internal TV system. These terabytes of images are analyzed by artificial intelligence to help assemble the dynamics of what happened and to identify those involved.

For identification, the system scans the videos and identifies, frame by frame, all faces found.

For this, the intelligence developed by the PF demarcates 512 points on each of the faces. These notes are compared with the faces identified by the same process in the photos of the prisoners taken at the time of detention, in government image banks, videos taken from cell phones and other corporate databases.

The intersection of this information points to the compatibility of the two sides. The system is so advanced that in recent days it has been possible, according to experts, to identify people with masks or with almost the entire face covered with cloths.

Although the artificial intelligence presents the results, the experts also do a manual post-work to confirm the results of the machine.

To increase the efficiency of the analysis, the experts took pictures at the beginning of the investigation, when the prisoners were registered, with images at closer angles than those taken by the closed circuit cameras.

In addition, as reported by investigators, the work has been facilitated by the quality of the images of the internal circuits.

Scenes from places where the damage was greatest and some specific ones, such as the breaking of Dom João’s clock and the depredation of Di Cavalcanti’s painting, are being prioritized.

The work of producing the report with all the dynamics of the attacks and identification of those involved is considered the biggest ever done by the corporation’s experts.

In addition to all the images collected, the PF itself made its own and captured it with a 3D camera at the crime scenes to use in the production of a video with the entire course of the attacks, depredations and displacements of the coup leaders.

The artificial intelligence used in recognition was developed from open source codes of software for facial identification.

Based on this information collected on the internet, the PF’s audiovisual sector adapted the system for the purposes of the police, who at that time were trying to solve the problem they faced with identifying images.

Two years ago, reported an expert in the sector, all the work was done manually and there was difficulty in using the internal TV circuits because of the low quality of image capture.

From the development of the tool, the experts also created a methodology for use with double checking. First the machine. Then, the professional specialist in identification.

The experts’ goal is to integrate an ever-increasing number of databases with artificial intelligence so that the identification capacity is greater.

Among the current obstacles, experts report the case of the Superior Electoral Court, which has a huge bank of images, but it is the TSE (Superior Electoral Court) itself that makes the comparison and gives the answer on identification.

The ideal, they say, would be for the INC to have access to the TSE bank so that the artificial intelligence itself can check based on its reading model.