A client company, expert in consultancy for the safety of construction and naval sites, was experiencing problems in tracking and identifying risk events in docks and construction sites.

The client company faced a series of problems related to the complexity and vastness of the sites to be monitored: the use of traditional control methods, such as manual video surveillance and paper checklists, proved inefficient and unreliable.
Controls on operators and vehicles, in a dynamic environment like a construction site, required a large amount of human and material resources, without guaranteeing certain and accurate results.

In this sense, in sectors such as construction, where workers are often in risky conditions, the adoption of innovative technologies such as AI and the Internet of Things can be of primary importance. These systems, in fact, guarantee constant monitoring and high precision control, promoting more punctual compliance with legal regulations and a more flexible protection network.

On the other hand, our partner had already tested automated video surveillance tools but without achieving the desired results.
Therefore, after a long phase of analysis, we decided to develop a solution as personalized as possible to the needs and peculiarities of the client's work context.

For the partner, we created a monitoring system integrated with devices and sensors connected to the network.
The system allows real-time tracking of operators' activities: it follows their movements, monitors the most at-risk positions through video analysis, and ensures that safety standards are met.

To achieve this result, our engineers worked on two partially parallel tracks. The AI development team, in fact, focused on researching and fine-tuning the most suitable monitoring and anomaly detection models; while the IoT developers dedicated themselves to integrating the system into the monitoring devices.

The AI team focused on developing software capable of recording and analyzing various types of data in real-time: images collected from security cameras and raw data from sensors worn by operators.

Various Computer Vision, Anomaly Detection, and Data Science algorithms were then implemented and integrated into a single software - to identify and possibly signal potential danger situations or anomalous behaviors in the recorded events.
The anomalies to be reported included operational errors, positioning in dangerous areas of the site, performing high-risk activities, or poor application of prevention rules.

The complexity and diversification of situations also made it necessary to adopt some more traditional practices already in use, such as checklists - renewing them through automatic verification functions.

While the Machine Learning specialists were working on implementing the analysis and detection system, the embedded team was engaged in developing monitoring and alert functions on physical devices and interfacing between the AI software and the devices.
In this way, constant communication between cloud functions and sensors was ensured, and continuous monitoring of each component was optimized.

After a long final testing phase (also carried out on-site, to verify the full reliability of the system), our partner adopted this new technology with all its clients, which quickly proved to be more reliable and effective in identifying risk situations compared to the previously used control systems.
Risk reports increased, and incidents and behavioral discrepancies significantly decreased.