With automated screening, the professional can supervise 1,000 to 2,000 cameras simultaneously
developed a solution to optimize the use of security cameras and transform thousands of hours of video into useful information. The tool, called Agatha, to learn behavior patterns in monitored environments and issue alerts when it identifies changes. The system converts passive cameras into active monitoring devices, discards images that do not require attention and allows a single professional to monitor thousands of screens without being overwhelmed by irrelevant notifications.
To create the solution, Rafael Libardi, founder of the startup, questioned the limits of traditional systems: why do security cameras still largely work as sophisticated but not very intelligent motion sensors? The platform’s logic was initially tested in the world of , when Libardi worked on a project for the Armed Forces of a Latin American country.
The objective was to identify anomalous behavior in internal computer networks to detect foreign intrusions into local digital infrastructure. To do this, the system used the recognition of unusual communication patterns on the network.
Libardi realized that the method could be applied to security camera images. It would be enough to replace data packets with pixels and cyber attacks with non-standard behavior.
“What existed on the market was basically movement detection and, for any deviation, there was an alert. This produced thousands of notifications per hour and made the system of little use”he stated. “I decided to combine what I knew in cybersecurity with visual security.”
Pattern Learning
Instead of operating based on fixed rules, the platform observes the environment for a certain period of time and establishes what it considers normal in that context: where vehicles are usually parked, which times are busier and which areas concentrate more traffic.
From this baseline, the system identifies deviations and forwards alerts for human evaluation.
Libardi cites studies on video surveillance to illustrate the limits of human attention. Research indicates that after approximately 12 minutes of continuous observation, an operator may miss up to 45% of the activity displayed. After 22 minutes, up to 95% of what happens goes unnoticed, even when few cameras are on display.
In this context, an operator can only monitor a few dozen cameras with quality before fatigue compromises surveillance. With automated screening, the professional can supervise 1,000 to 2,000 cameras simultaneously, because he only receives the sections that require analysis.
According to the startup, the tool filters out more than 99.8% of irrelevant images and directs the analyst’s attention to situations that require decisions. “He only sees what is strange”Libardi summarized.
Applications beyond security
The number of cameras installed is growing in condominiums, companies, public roads and events. The (Brazilian Association of Electronic Security Systems Companies) estimates that the sector earned R$14 billion in 2024, an increase of 16.1% compared to the previous year.
In addition to real-time monitoring, the technology enables forensic analysis, allowing you to review large volumes of video and quickly locate moments when there was an anomaly.
An energy distributor in Minas Gerais, for example, faced recurring thefts at substations and needed to review weeks of recordings after each incident. With Agatha, hours of video were reduced to around 10 minutes containing the exact moment of the invasion.
The solution has also been used to verify the correct use of PPE (personal protective equipment), detect behaviors that precede work accidents, control stocks in warehouses and monitor industrial processes.
In an agro-industrial industry in Belém (PA), the technology started to identify signs of wear in large chains through subtle changes in visual patterns, such as atypical vibrations, irregular inclinations and changes in the texture of the components.
Another project used the tool to perform real-time counting of bags of cement, feed and grains at the Port of Santos, replacing manual processes prone to errors.
Use in smart cities
The tool’s adaptation time varies depending on the application. PPE monitoring can be implemented in less than 24 hours, while bag counting systems at ports require around a week of adjustments. More specific industrial applications may require months of .
In monitored condominiums and neighborhoods, AI identifies residents’ vehicle license plates and alerts when unknown cars remain nearby for periods considered unusual.
In one specific case, the system issued an alert when it detected a child near an automatic garage door during opening. The combination of factors was classified as anomalous and allowed the operator to stop the mechanism in time.
The technology can also be used at major events and complement smart city initiatives. Programs like Smart Sampa, in the capital of São Paulo, use facial recognition to locate suspects. The Noleak platform can act in a complementary way by focusing on the behavioral analysis of images.
Experts highlight that facial recognition technologies have margins of error and require human validation. Libardi agrees.
“No solution should operate in isolation. They work as precision filters. There must always be subsequent verification because the tool does not replace the human eye; it reorganizes priorities”these.
The researcher also highlights the importance of adequate infrastructure. “It is necessary to educate the customer because technology is not magic. The camera needs to be in the right place and have reasonable image quality. Sometimes, the customer believes that they will be able to identify something 200 meters away with a low-cost camera”he stated.
With support from FAPESP’s PIPE (Innovative Research Program in Small Businesses), the startup restructured its data architecture and improved mathematical models to gain scale.
“The researcher masters the technique, but does not always know how to transform it into a commercially viable product”declared Libardi.
This text was by Fapesp Agencyon May 2, 2026. The content is free for republication, the source is cited, and was adapted to the Poder360 standard.
