cross
  • About us
  • Blog
  • NEWSLETTER
    • Request a demo

    Computer vision for workplace safety

    Machine vision applied to safety involves using cameras and image sensors combined with analytical algorithms, often based on machine learning, to automatically detect hazardous situations in the workplace. These situations include the presence of people in dangerous areas, intrusion into machine or vehicle traffic zones, lack of personal protective equipment, unsafe postures or behaviors, abnormal facility conditions, obstacles in evacuation routes, spills, smoke, or fire. The results can trigger safety functions (stopping a machine, reducing the speed of a robot or vehicle, blocking an access point), generate real-time alerts for supervisors and operators, or provide data for indicators and reports used in preventative analysis. A distinction is made between vision systems certified as sensitive protective devices, which are part of machine safety functions according to IEC 61496 and ISO 13849 standards, and non-certified video analytics systems, which provide alerts and information but cannot replace the protections required by machinery legislation. In Spain, their implementation is subject to Law 31/1995 (risk assessment, consultation and participation in the face of new technologies), Royal Decree 1215/1997, data protection regulations (General Data Protection Regulation and Organic Law 3/2018, in particular its article 89 on workplace video surveillance) and, when they incorporate artificial intelligence, to Regulation (EU) 2024/1689, which considers systems intended to monitor and evaluate workers to be high risk and prohibits the recognition of emotions in the workplace except for medical or safety reasons.

    In short

    The use of cameras and image analysis algorithms, often incorporating machine learning, to automatically detect risk situations (intrusion into hazardous areas, lack of protective equipment, traffic, falls, postures, smoke, or spills) and activate safety functions, alerts, or indicators. Only devices certified according to IEC 61496 and ISO 13849 can perform machine safety functions; non-certified video analytics systems are an additional layer. Their implementation requires risk assessment and consultation (Law 31/1995), compliance with data protection and workplace video surveillance regulations (GDPR, Organic Law 3/2018, Article 89), and, when they incorporate artificial intelligence that monitors people, compliance with the high-risk obligations of Regulation (EU) 2024/1689, which also prohibits the recognition of emotions in the workplace except for medical or safety reasons.

    Content
    1. Applications and types of systems
    2. Legal and data protection requirements
    3. Design and implementation criteria
    4. Organizational application: how to implement machine vision successfully
    5. Limits and common mistakes
    6. Practical example
    7. Regulatory and reference framework
    8. Related concepts
    9. References

    A–Z dictionary →

    Applications and types of systems

    The most widespread applications are intrusion detection in dangerous areas of machines, robots and vehicles, with activation of safety functions or alerts; verification of the use of personal protective equipment (helmet, goggles, vest, harness) in access points and controlled areas; detection of people in the vicinity of forklifts and industrial vehicles, with warning to the driver or speed reduction; monitoring of traffic and the separation between pedestrians and vehicles; detection of falls, stationary people or lone workers; ergonomic analysis of postures and movements from video; detection of smoke, fire, leaks and spills; and verification of order, signage and evacuation routes.

    From a technical and legal perspective, it is useful to distinguish three categories. Certified optoelectronic protective devices (photoelectric curtains and barriers, safety laser scanners, safety vision systems compliant with IEC 61496) are designed, validated, and classified to be part of machine safety functions with a specific performance level, and can replace physical guards under the conditions stipulated by the risk assessment. Uncertified AI-powered video analytics systems detect situations and generate alerts or logs, but their reliability is not guaranteed for safety functions and they can only be used as an additional layer of protection. Finally, systems that monitor and evaluate the behavior of specific individuals fall under the scope of high-risk systems in Regulation (EU) 2024/1689 and workplace video surveillance, with enhanced requirements.

    The European Agency for Safety and Health at Work, in its line of work on digitalization, analyzes intelligent safety and health monitoring systems (cameras, wearable devices, drones, smart glasses) and points out both their potential to remove people from dangerous tasks and detect risks, as well as the privacy problems, psychosocial risks associated with the feeling of surveillance and technological dependence that must be managed.

    Legal and data protection requirements

    • Risk assessment and consultation. Implementation is a new technology that requires consultation with worker safety representatives (Article 33 of Law 31/1995) and review of the risk assessment, including psychosocial risks.
    • Machine safety functions. Only devices certified according to IEC 61496 and ISO 13849-1 can assume safety functions; integration requires risk analysis, positioning according to ISO 13855 and validation.
    • Workplace video surveillance. Article 89 of Organic Law 3/2018 allows the processing of images for workplace monitoring with prior, express, clear and concise information to workers and their representatives, and prohibits cameras in rest areas, changing rooms, toilets and similar places.
    • Principles of data protection. Lawfulness, minimization, purpose limitation, limited storage, impact assessment when there is systematic observation and recording of processing activities.
    • Artificial intelligence regulation. Artificial intelligence systems intended to monitor and evaluate the performance and behavior of workers are high-risk (Annex III), with obligations regarding risk management, data quality, human supervision, transparency and registration; emotion recognition at work is prohibited except for medical or safety reasons; most of the obligations for high-risk systems are applicable from 2 August 2026.
    • Digital rights. Article 20 bis of the Workers’ Statute recognizes the right to privacy against the use of video surveillance and geolocation devices.
    • No covert disciplinary use. A system implemented for preventive purposes cannot be used for disciplinary control purposes without meeting the requirements specific to that purpose, and misuse violates data protection regulations.

    Design and implementation criteria

    • Defined preventive purpose. Specific objectives (intrusion detection, equipment use, traffic) linked to assessed risks and prior technical and organizational measures.
    • Privacy by design. Edge processing without image storage, anonymization or blurring of people, minimal retention, and restricted access.
    • Reliability. Testing under real-world conditions (lighting, occlusions, clothing, weather), acceptable false positive and negative rates, and human monitoring of alerts.
    • Integration with safety functions. Use of certified devices for safety functions and analytics systems as an alert and information layer.
    • Transparency. Clear information to staff and representatives about what is detected, for what purpose, who has access and for how long the data is kept.
    • Psychosocial factors. Evaluation of the effect of perceived surveillance and design oriented towards collective protection, not individual control.
    • Cybersecurity and maintenance. Protection of cameras and servers, model updates, and periodic operational verification.
    • Governance. Defined responsibilities, alert management procedure, results review and system improvement.

    Organizational application: how to implement machine vision successfully

    1. Define the preventive purpose and the risks to be controlled, and verify that the prior technical and organizational measures are in place.
    2. Consult the project with the worker safety representatives and the health and safety committee, and carry out the data protection analysis (purpose, minimization, retention, impact assessment) and, if the system incorporates high-risk artificial intelligence, the analysis in accordance with Regulation (EU) 2024/1689.
    3. Decide on the architecture: certified devices for safety functions and analytics systems as an alert layer; design with privacy from the design stage (local processing, anonymization, no cameras in rest areas or changing rooms).
    4. Inform the staff in advance, expressly, clearly and concisely about the system, its purpose, access and maintenance, and mark the areas.
    5. Test the system under real-world conditions, measure its reliability, establish human oversight of alerts and response procedures, and train managers and operators.
    6. Integrate alerts and indicators into the preventive management system for analysis and for review of risk assessment, without covert disciplinary use.
    7. Periodically review the effectiveness, false positives and negatives, psychosocial effects and regulatory compliance, and update the system.

    Preventive management software allows you to receive and integrate alerts and indicators generated by vision systems, link them to areas, equipment and risk assessments, manage the resulting actions and maintain the traceability and information and consultation records required by the regulations.

    Limits and common mistakes

    1. Using non-certified video analytics systems as a substitute for guards or protective devices required by machinery legislation.
    2. Installing cameras without prior information to the staff and their representatives, without data protection analysis, or in prohibited areas.
    3. To divert the preventive purpose towards disciplinary control or performance evaluation without the corresponding guarantees.
    4. Relying on systems with unmeasured error rates or without human supervision of alerts.
    5. Ignoring the psychosocial risks of perceived surveillance.
    6. Not considering the obligations of the Artificial Intelligence Regulation for high-risk systems or the prohibitions (emotion recognition).

    Specific data protection, artificial intelligence and machine safety requirements must be analyzed on a case-by-case basis; this document is for informational purposes only.

    Practical example

    Situation: A logistics center with heavy forklift and pedestrian traffic is experiencing several close-range incidents and wants to implement an artificial vision system to reduce them.

    • Design. The prevention service, systems, the data protection officer and the workers’ representatives define the purpose (detection of pedestrians in maneuvering areas and warning to the driver) and the architecture: certified laser scanners on the forklifts for speed reduction and cameras with edge analysis, without image storage or identification of people, for alerts at crossings.
    • Guarantees. A data protection impact assessment is carried out, staff are informed beforehand and clearly, the areas are marked, rest areas are excluded, and it is documented that the system is not used for disciplinary or performance evaluation purposes.
    • Implementation. The system is tested for one month by measuring false positives and negatives, the zones are adjusted, and drivers and managers are trained in responding to alerts; the aggregated alerts are integrated into the preventive management system.
    • Results. Proximity incidents are clearly reduced in six months, the analysis of alerts by zone allows for the redesign of two intersections, and the periodic review confirms the absence of relevant psychosocial effects and regulatory compliance.

    Regulatory and reference framework

    The IEC 61496, ISO 13849-1 and ISO 13855 standards regulate optoelectronic protection devices and their integration into safety functions, and the Spanish Data Protection Agency publishes criteria on workplace video surveillance.

    Related concepts

    References

    1. Official State Gazette. Law 31/1995, of November 8, on Occupational Risk Prevention. 1995, current consolidated text. Official source
    2. Official State Gazette. Royal Decree 1215/1997, of July 18, establishing the minimum health and safety requirements for the use of work equipment by workers. 1997, current consolidated text. Official source
    3. European Union. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data. 2016. Official source
    4. Official State Gazette. Organic Law 3/2018, of December 5, on the Protection of Personal Data and the guarantee of digital rights. 2018, current consolidated text. Official source
    5. Official State Gazette. Royal Legislative Decree 2/2015, of October 23, approving the consolidated text of the Workers’ Statute Law. 2015, current consolidated text. Official source
    6. European Union. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules in the field of artificial intelligence. 2024. Official source
    7. European Agency for Safety and Health at Work. Digitalization of work. Thematic portal. Official source

    Editorial information

    Publication date: August 30, 2026 .

    Editorial Manager: Sabentis Editorial Team .

    Editorial review by Pablo Rodríguez LinkedIn

    Executive Vice President of the ORP International Foundation and Chief Financial Officer of Sabentis.

    Request a Demo

    Discover all that Sabentis can do for your organization.

    Try Sabentis

    request a demo
    stars 5
    GetApp Software Advice Capterra