Video analytics for control of industrial safety rules and regulations: “out-of-the-box” solutions versus adaptable systems

It is quite common that clients would like to obtain out-of-the-box perfectly functioning ready-to-use solution. Unfortunately, for the applications of industrial security and labour safety this is rather impossible. Huge variety of tasks, uniqueness of industrial facilities great variability of the camera angles, lighting conditions and control areas – all of this prevents out-of-the-box solution to show high performance under field conditions. As a result, we find that new adaptable systems most often replace the out-of-the-box solutions.

Out-of-the-box solutions: major limitations

The baseline functionality for the most of out-of-the-box solutions focuses on the monitoring of personal protective equipment (PPE) use by the personnel. However, most often, the variety of tasks for industrial control of safety is much more diverse. It is undeniable, however, that out-of-the-box solutions are much more attractive in terms of price and ease of implementation, but it must be noted that these solutions most often possess number of limitations that customer should be aware of. These limitations might include:

1. Low recognition accuracy. The high recognition accuracy of out-of-the-box solutions is most often achieved only under ideal controlled conditions for specific camera angles, lighting conditions, etc. For the cases with challenging industrial conditions with high concentration of dust in the air, high temperatures, poor lighting conditions, large variety of possible camera angles the out-of-the-box solutions most often reach the recognition accuracy of 50-60%, which is not enough for the tasks where the main objective is to ensure safety of people.
2. Limited baseline functionality without the support for further customisation. Most often out-of-the-box solutions have relatively limited functionality and do not support further customisation and expansion with additional functionality options.
3. Absence of individual customisation to the conditions of facility. Most of out-of-the-box solutions do not provide adaptation of the system to accommodate existing processes and conditions at the client’s facility.

As a result of the reasons listed above, the out-of-the-box solutions most often lose their appeal, especially in challenging and complex environments. We see that clients choose adaptable solutions more often, and these systems enable to have high recognition accuracy, address unique tasks at production facilities, provide flexible configuration of the software and enable convenient integration with access control systems, solutions for industrial system automations, etc.

Adaptable solutions: smart customisation

First of all, let us define the concept of adaptable solutions. Adaptable solutions are presented in form of baseline software that can be easily developed further to meet specific requirements of the facility and be directly integrated with existing production processes. At the same time, however, such software has quite extensive baseline functionality which can be extended further for addressing new, specific and unique tasks at the facility.

For instance, EYECONT system, developed by Mallenom Systems, includes 12 modules for video analytics including the control for 15 PPE types; detection of people in restricted and hazardous production areas; detection of smoking in smoke-free zones; identification of smoke and open fire, and left unattended objects; control of loading and unloading operations; and more.

User interface of adaptable EYECONT system for video analytics

At the same time, if the client requires to address such task as control of the actions performed by the personnel for a stage of production process, such functionality can be added based on the existing software quite efficiently.

Adaptation of software solution to unique properties of industrial environments is sometimes the only way to deliver efficient and quality product. Such adaptation may include further training and fine-tuning of neural network models. The ability to perform further adjustment of the neural networks allows to increase the accuracy of industrial accidents detection up to 98%, which in turn allows to promptly respond to detected accidents and safety violations.

Further development and adaptation of baseline software consists of the following steps:

1. Collection of visual data from the implementation site. Video cameras that are installed at the facility are utilised to record video files for their further use during training of neural network models. To improve the time that is required for data collection, the imitation of the accidents and safety violations may be implemented. The more stable operation of the algorithms requires video images from multiple camera angles with different lighting and weather conditions. The more diverse the collected sample the less false positive activations the system would have.
2. Data makeup of selected samples for training of neural networks. After receiving sufficient amount of data, the data engineer together with enterprise occupational safety officer makeup the data. The data makeup consists of manual identification and classification of violations for the further training of the models.
3. Further training of neural network models given the unique conditions of the implementation environment. Prepared data is input into trained neural networks for the new training stage which improves quality of operation of the model in set conditions. Particularly, this additional training stage helps to improve classification accuracy and detection stability, and decrease number of false positive activations.
4. Integrations of newly trained model with the software. Newly trained neural network model is connected to the existing software and supplied to the client.
5. Testing of the software and on-site warranty testing. The final stage of software adaptation consists of final system testing which examines whether the system meets the originally set requirements.

Adaptability of the system allows to the client to receive solution which:

• is highly accurate in monitoring industrial safety rules violations 24/7 and allows to collect evidence of the recorded accidents;
• enables to address generic and unique tasks as a part of single system;
• enables to perform integration with production and business procedures of the enterprise easily;
• accommodates further flexible training of the system in case of change in any of the processes at the facility.

Therefore, given the crucial role of high recognition accuracy in detection of industrial accidents and violations in industrial safety rules and regulations, the customisation and adaptation of the software most often is simply necessary. Moreover, adaptability of the solution helps the client to flexibly incorporate the control system for labour health and safety with already existing systems and set conditions of the facility.

Anna Solovyova, Alena Venediktova
Published in «System Safety»