ITCL has completed the In-Lab tests of the body pose estimation module. Image capturing processes have been carried out on 10 workers from the company who are over 45 years of age, performing different tasks in two separate simulated workplace environments, office and manufacturing.

Within the simulated manufacturing environment, workers were asked to assemble a pile of boxes and then disassemble it (see Fig. 1), both standing and seated, while the images were captured from the side (see Fig. 2) and from behind (see Fig. 3). For the simulated office environment, they were asked to write a text on a computer while images were taken from one side (see Fig. 4). These tasks were thought up in order to force the participants to adopt a range of ergonomic and non-ergonomic postures. For example, the chair was situated at different distances to the table during the tests simulating the office environment.

Figure 1. Boxes assembly explanation for the image capturing process in the manufacturing environment.

Once the images were captured, they were processed by the posture estimation program where the workers’ joints are obtained (skeleton in green in figures 2, 3 and 4). From these results it will be easy to calculate if the worker’s posture is adequate for the task, or on the opposite side, that those positions of the joints can be harmful over time.

Figure 2. Manufacturing environment tests from the side.

Figure 3. Manufacturing environment tests from behind.

Figure 4. Office environment test from the side.

We would like to thank all the ITCL employees who spent some of their valuable time to participate in these tests.

Telespazio has finalised the development of the Location and Emergency service of the WAOW tool. The in-lab tests have proven the capacity of the service to provide the user’s location in both indoor and outdoor environments. In addition, the Emergency service, called E112 and based on the implementation of the PEMEA standard, has been tested successfully. The development of two additional servers for the testing of the PEMEA architecture (PSAP side of the PEMEA protocol) has proven very useful. The integration phase is now ongoing and will allow a complete testing of the E112 service during the Short-term tests. This test will involve other partners’ sensors for the monitoring of health data and the Decision Support System responsible for the automatic triggering of the E112 by the WAOW tool.

The next step for Telespazio is to supervise the deployment of the WAOW tool in one of the pilot site covering the use case ‘Office’: Mutua Universal. Mutua Universal is a collaborative mutual society based in Barcelona. This deployment raises many challenges. Some of these challenges are due to the distance between the Telespazio team (in Toulouse, France) and Mutua, like for instance remote organisation of the deployment or unfamiliarity with the offices. Other challenges will be faced by all partners responsible for the deployment and are intrinsic to the WAOW tool’s nature: the large-scale deployment of dozens of sensors, routers and computers in offices, the integration of the WAOW tool into the company network while respecting all security policies and so on.

However, as demonstrated since the very beginning of this project, the strong collaboration between partners will help this deployment being a success!

The In-Lab experimental phase of the WorkingAge project is completed. The tests aimed at identifying and validating the set of sensors and methodologies for developing the WorkingAge of Wellbeing (WAOW) tool. Such experiments were conducted in several European countries, corresponding to the partner’s premises, challenging the difficulty of running experimental protocols during the COVID-19 pandemic. All the experiments were run in compliance with all the social distancing practices and hygiene standards stated by the World Health Organization (WHO).

The WorkingAge In-Lab tests were based on a shared experimental protocol between partners of the consortium. This aspect allowed us to enlarge the experimental sample and reach a greater robustness of the experimental outcomes. In this regard, BrainSigns recruited 17 participants and tested different solutions for acquiring the Electroencephalogram (EEG), Electrodermal Activity (EDA) and Electrocardiogram (ECG) data with the aim to identify the best technical combination of sensors to be integrated in the WAOW tool.

The shared experimental protocol included three different tasks for the participants: such tasks were selected to modulate the mental workload, stress, and emotional state of the participants, and to simulate the daily working activities whose will be carried out by the workers during the next experimental phases (In-Company tests) in realistic working environments.

The next step will consist in the validation of the data analysis algorithms and integration of the DSS modules which will be used in the final version of the WAOW tool.

EXUS recently completed the in-lab testing of its Facial Recognition and Authentication and Gesture Based Interaction platform, as developed for the WorkingAge project. Despite complications arising from the COVID-19 pandemic involving limitations on the number of personnel allowed to be present simultaneously in our offices and restrictions on movement within the city, we “recruited” 10 people to run the system through its paces and discover how easy (or hard!) they found it to learn and interact with it.

First, a few words about the system itself. The component allows registered users to interact with the WorkingAge Tool with hand gestures. A user registers simply by uploading their selfie, a step requested when registering on the mobile app. For our in-lab tests, users provided their profile picture from the company online registry. Any gesture can then be programmed to correspond to a defined interaction with the system. The gestures could be enumeration with fingers, the “thumbs up” or “ok” gestures, or any other combination desired. Such an interaction platform has a wide range of applicability, from enabling non-verbal communication for law enforcement agents, a means to digitize and automatically record gestures performed by referees in sporting events, to workers interacting with a platform designed to improve their health, as in the WorkingAge project. To achieve system functionality, a suite of Artificial Intelligence and Machine Learning techniques was implemented, relying primarily on deep learning employing convolutional neural networks.

Through our in-lab tests and the experience of our participants, the users found the gesture-based interaction system intuitive and easy to use. Sample system output from two users is shown in the picture. A “similarity score” is returned for each user; here results of 67-69% are achieved, which are sufficient to correctly identify and authenticate the user and expected values based on the user-camera distance. Also shown is the enumeration gesture value the system recognized, correctly identifying the gesture in each case. The tests revealed some surprises as well, namely that the system worked better with a solid (white) background, and that it had difficulty correctly identifying gestures 4 and 5. This valuable feedback will be used to improve the system in future iterations.

We had a very pleasant experience testing our system on real users in our controlled environment, and with the in-lab tests of all partners nearing completion, we are looking forward to undertaking the integration of all components into the complete WorkingAge Tool prototype!

The WorkingAge team has been working hard in recent months to complete the development of the WorkingAge Of Well-being (WAOW) Tool. In parallel, all the partners, both industrial partners and academic partners, are defining the individual exploitation plans for the WAOW tool to ensure the sustainability of the project’s results beyond the project end and to demonstrate how WorkingAgehas influenced the EU landscape.

One of the first steps in the definition of the dissemination and exploitation results strategies is the identification of the target audience, not only for the WAOW tool but also for all the results and achievements of the WorkingAge project.

• Academia, scientific/research organisation: to this target group belongs universities or research centers in the fields such as Silver Economy, Healthy Ageing, Smart manufacturing, Assisted living, Social Science, Cybersecurity, Psychology, ICT and Data Science.

They can benefit from the project research findings, such as scientific articles, reports, conferences and scientific events. So that synergies and collaborations can be established between research lines of different entities and the WorkingAge project.

• Commercial Players and Investors: to this target group belongs commercial entities or investors of fields of voice and gesture recognition, Neurometrics, Eye recognition, Robotics, data management, Apps developers, Training Centers, Manufacturers of telephones, mobiles, tablets, smartwatches, virtual glasses, R&D Department of Medical Universities, Insurance companies, etc.

They expect in-depth project information and in particular the project method, design approach, interactions designed and concept, general information about WAOW Tool and the business model.

They can benefit from an innovative and pioneering tool that can generate financial benefits when it will be launched on the market. In this way, marketing and sales alliances can be established between commercial actors and investors and WorkingAge consortium.

• Final Users: to this target group belongs workers and companies.

They expect brief information about the WorkingAge project and in-depth information about the WAOW Tool, the validation tests, the pilot test results, the benefits and advantages of the use of WAOW Tool, the regulations or different training materials.

They can benefit from the use of the WAOW Tool that allows the improvement of the health and well-being of people at work by supervising their working conditions and providing different types of advice through personalized technologies and friendly & intelligent human interfaces. With the use of the WAOW Tool, the occupational health and safety of workers can be strengthened, but not only physical health, also mental health, which is one of the major aspects neglected by companies.

• International, European, national and local associations, Policy makers, Standards associations: Their interest lies mainly in the possibilities for Healthy Ageing and work related improvements, also with the products and services that the WAOW Tool can provide and the standardization or regulation processes related to the outcomes.

They can benefit from the WorkingAge research findings, which can be potentially relevant to provide recommendations and good practices on new methodologies, new intervention procedures, new possible products, economic aspects, new standards and new regulations.

• General audience: General public aware of the effort the EU is doing regarding research and development and interested in new opportunities for European Enterprises.

They can benefit from the basic information about the WorkingAge project, its activities and its outcomes and achievements, to understand its key aspects and take advantage of them for the general benefit of society.

Interested in have a said in the WORKINGAGE innovation process?

You can subscribe our newsletter or contact us (info@workingage.eu) to know opportunities for stay informed or for participating.

If you want to know more about the progresses of WorkingAge project and the developments of WAOW Tool, follow our social networks (Twitter: https://twitter.com/Workingage_EU & LinkedIn: https://www.linkedin.com/in/workingage-EU) and our blog (https://www.workingage.eu/category/blog), where WorkingAge periodically uploads information about the project and its achievements.

In a previous description of the scope of the E112 Service in WorkingAge, we mentioned that the emergency message generated by the WAOW tool and sent to a PSAP, will also include medical data of the WAOW user, after explicit consent has been provided. This feature enables the study of the impact of medical data provision on the emergency call handling, the situational awareness of first responders and the resulting planned emergency response.

In the article about “How can PSAPs benefit best from enriched data sending about an emergency event?”, we raised the questions:

• How to display these data to make it easy to use by the emergency service operator?
• How to represent medical data?
• What weight should be given to such data in the decision making?
• Should these medical data be transformed into a list of required medical equipment?
• Should this information be used to dimension the medical teams’ size?

At least 40 emergency applications for smartphones provide some kind of medical data to PSAPs and emergency call-takers. But how useful is this data?

In September 2020, EENA published the report “Impact of COVID-19 on PSAP activities”. The report describes the challenges that the COVID-19 pandemic imposed on emergency services and the required adaptations taken in the short time frame. In the second section describing how PSAPs handled the impact of COVID-19, EENA surveyed 32 emergency services professionals from 25 countries, between 27 May to 30 June 2020. Responses were received from of police, fire and rescue services, emergency medical services, and 112 PSAPs / Emergency Management Agencies.

One of the questions in the survey asked, “Would it have been useful to automatically receive basic medical data?”

• 59% of respondents stated automatic reception of basic medical data would have been useful during the COVID-19 outbreak
• 20% of respondents stated it wouldn’t have been useful
• 22% of respondents were unsure

Respondents were overall supportive of introducing the opportunity for call-takers/dispatchers to automatically receive medical data about callers, including previous medical conditions, which might have particular interest during the COVID-19 outbreak.

Figure: Responses to the question “Would it have been useful to automatically receive basic medical data?” [Source: EENA, https://eena.org/knowledge-hub/documents/report-impact-of-covid-19-on-psap-activities/, p. 43]

The RWTH Aachen University, located in Aachen, Germany, is one of the Universities working on the WorkingAge Project.

The University was founded in 1870 and first received the name “Königliche Rheinisch-Westphälische Polytechnische Schule” (Royal Rhenish-Westphalian Polytechnic School) which was later “shortened” to “Rheinisch-Westfälische Technische Hochschule Aachen” (Rhine-Westphalian Technical University Aachen).

With more than 45,000 students, the RWTH Aachen University is the largest University for technical studies in Germany.  

This year the University is turning 150 years old and in order to show the many things the RWTH Aachen University is working on, has worked on and achieved over the years, a “150 Years of RWTH” Special was filmed and uploaded onto YouTube, the link can be found down below. Even though the video is in German, English subtitles can be chosen in the settings.

The actual department, which is working on the WorkingAge Project, is the “Institute of Industrial Engineering and Ergonomics”. This institute is the oldest Institute in Germany dedicated to this field of research and was founded in 1928. Buzzwords such as Industry 4.0, demographic change and digitalisation increase the relevance of ergonomic workplace design to enable a long-term healthy working life and ensure the protection of the safety and health of employees. Against the background of such a variety of influencing factors, their interaction and dynamics, which lead to a constant change in work, the need for interdisciplinary research to ensure humane and economic efficiency as well as effective work is becoming increasingly important. Therefore, the aim of the WorkingAge project is more important than ever to guarantee a long-term healthy working and private life. We are therefore pleased to be part of the WorkingAge consortium and to be able contribute to healthier working conditions around Europe.

Link to the “150 years of RWTH Video”

We are at the verge of the WAOW Tool deployment phases and Green Communications (GC), that is developing through the WorkingAge project and other initiatives a new generation of open IoT platform, is defining its exploitation plan.

Our proposition is to combine cutting edge M2M communication technologies with edge cloud and services. Thus, to increase bandwidth, servers’ response time, mobility, energy saving, and data sovereignty of massive IoT, Internet of Moving Things and other professional IoT applications.

The market opportunity for GC’s new solution has been estimated at 8 billion of euros and 450 million IoT connections in Healthcare, Manufacturing, Transportation, Governments, and Smart Cities sectors in Europe.

In the field of prevention of work-related diseases, the participation of INTRAS in the European Working Age project stands out. This project studies the profile of workers over 50 years of age and the promotion of healthy habits in the work environment and in activities of daily life, and thus improve working and living conditions.

In relation to awareness, INTRAS is constantly observing good practices and healthy habits and promoting actions to create, disseminate or improve a culture of occupational risk prevention.

INTRAS participates in or leads other projects and initiatives to promote healthy habits, non-sedentary lifestyles and active aging, such as the constitution of the “Club Deportivo Duero”, a standardized and inclusive club that promotes the practice of physical activity and sports and the transmission of the physical, mental and social benefits of this activity, with values such as companionship, teamwork, communication, active listening, commitment, responsibility or punctuality.

Recognition of expressions of emotions and affect from facial images is a well-studied research problem in the fields of affective computing and computer vision with a large number of datasets available containing facial images and corresponding expression labels [1]. Thanks  to  the  unprecedented  advances  in  machine  learning  field,  many  techniques  for  tackling  this  task  now use deep learning approaches which require large datasets of facial images labelled with the expression or affect displayed. An important limitation of such a data-driven approach to affect recognition is being prone to biases in the datasets against certain demographic groups. The datasets that these algorithms are trained on do not necessarily contain an even distribution of subjects in terms of demographic attributes such as race, gender and age. Moreover, majority of the existing datasets that are  made  publicly  available  for  research  purposes  do  not  contain  information regarding these attributes, making it difficult to assess bias, let alone mitigate it.  Machine  learning  models,  unless  explicitly  modified,  are  severely  impacted by such biases since they are given more opportunities (more training samples) for  optimizing  their  objectives  towards  the  majority  group  represented  in  the dataset. This leads to lower performances for the minority groups, i.e., subjects represented with less number of samples.

To address these issues, many solutions have been proposed in the machine learning community over the years, e.g. by addressing the problem at the data level with data generation or sampling approaches, at the feature level using adversarial learning or at the task level using multi-domain/task learning. Bias and mitigation strategies in facial analysis have attracted increasing attention both from the general public and the research communities. Many studies have investigated bias and mitigation strategies for face recognition, gender recognition, age estimation, kinship verification and face image quality estimation.  However, studies specifically analysing, evaluating and mitigating race, gender and age biases in affect recognition have been scarce.

Therefore, in our recent research work [1] as part of the WorkingAge project, we undertook a systematic investigation of bias and fairness in facial expression. This is an important undertaking in the context of WorkingAge project because the average age of target user group is 45+, i.e. older than the average age of the datasets that are typically used for training predictive models for facial affect analysis. Therefore we need to understand whether and how the scarceness of expression data from older populations will impact our facia affect analysis module and the predictions it generates for the WAOW tool.

To this end, we considered three different approaches, namely a baseline deep network, an attribute-aware network and a representation-disentangling network under the two conditions of with and without data augmentation. Data augmentation refers to techniques used to increase the amount of data by adding slightly modified copies of already existing data or newly created synthetic data from existing data. In our attribute-aware solution, we provide a representation of the attributes as another input to the classification layer. This approach allows us to investigate how explicitly providing the attribute information can affect the expression recognition performance and whether it can mitigate bias. The main idea for the disentanglement approach is to make sure the learned representation does not contain any information about the sensitive attributes of race, gender and age.

With these three methods in place, we conducted experiments on RAF-DB [4] and CelebA [5] datasets that contain labels in terms of gender, age and/or race. To the best of our knowledge, ours is the first work (i) to perform an extensive analysis of bias and fairness for facial expression recognition, (ii) to use the sensitive attribute labels as input to the learning model to address bias, and (iii) to extend the disentanglement work of [3] to the area of facial expression recognition in order to learn fairer representations as a bias mitigation strategy.

The technical and experimental details of our study are provided in our research paper [1] and have been presented by Tia Xu at the of 16th European Conference on Computer Vision (ECCV 2020) Workshop on Fair Face Recognition and Analysis – http://chalearnlap.cvc.uab.es/workshop/37/description/.

In summary, what we found is that (i) augmenting the datasets improves the accuracy of the prediction model, but this alone is unable to mitigate the bias effect; (ii) both the attribute-aware and the disentangled approaches equipped with data augmentation perform better than the baseline approach in terms of accuracy and fairness; (iii) the disentangled approach is the best for mitigating demographic bias; and (iv) the bias mitigation strategies are more suitable in the existence of uneven attribute distribution or imbalanced number of subgroup data.

Below we provide a snapshot of Table 5 from our paper [1] to illustrate the challenging cases and qualitative results from the different models we have investigated.

The most relevant finding for the WorkingAge project is that, for the age group of 40-65, data augmentation, attribute awareness and disentanglement help improve the accuracy of the machine learning models for facial expression recognition, and the disentangled approach provides fairer classification results. We will take these findings into account when finalising our facial affect analysis module to be integrated into the WAOW Tool.

References

[1] Sariyanidi,  E.,  Gunes,  H.,  Cavallaro,  A.:  Automatic  analysis  of  facial  affect:A  survey  of  registration,  representation,  and  recognition.  IEEE  Transactionson  Pattern  Analysis  and  Machine  Intelligence37(6),  1113–1133  (June  2015).https://doi.org/10.1109/TPAMI.2014.2366127

[2] Xu, T., White, J., Kalkan, S. & Gunes, H. Investigating Bias and Fairness in Facial Expression Recognition, Proc. of  16th European Conference on Computer Vision (ECCV 2020) Workshop on Fair Face Recognition and Analysis, 2020. https://arxiv.org/abs/2007.10075

[3] Locatello, F., Abbati, G., Rainforth, T., Bauer, S., Scholkopf, B., Bachem, O.: Onthe fairness of disentangled representations. In: Advances in Neural InformationProcessing Systems. pp. 14611–14624 (2019)

[4] Li, S., Deng, W., Du, J.: Reliable crowdsourcing and deep locality-preserving learning for expression recognition in the wild. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2852–2861 (2017)

[5] Liu, Z., Luo, P., Wang, X., Tang, X.: Deep learning face attributes in the wild. In: Proceedings of International Conference on Computer Vision (ICCV) (December2015)