Since the COVID-19 outbreak in early 2020, body temperature monitoring has been adopted at the public entrances in many countries. In response to the global health crisis, CITIC Telecom CPC has designed and developed a “Thermal Imaging Solution” for measuring the body temperature of employees and visitors at its local offices.
On the one hand, the development of this smart temperature monitoring system was to implement the prevention and control measures of pandemic at the request of local governments to ensure the safety of the workplace and prevent the spread of the coronavirus in the community. On the other hand, CITIC Telecom CPC has a strong Data Science and Innovation Team from difference regions with advanced technologies. We hope to help businesses fight against pandemic with effective measures to resume operations as soon as possible.
The prototype of the “Thermal Imaging Solution” adopts industry-standard devices featuring both thermal imaging and general cameras. The system embedded with an alert function to provide a more accurate and effective body temperature measurement. To strive for excellence, the team has optimized the system with the previously-developed AI (Artificial Intelligence) models to provide a series of intelligent functions on top of body temperature measurement, including identity verification and mask detection.
Here we are honored to share with you how the AI functions were implemented in three distinct phases.
Phase 1: Data Collection:
To train the mask detection model, we first built a dataset consisting of 1,600 photos of people wearing masks and 1,600 ones without masks. In addition, the MS-Celeb-1M dataset was employed. It is an image library containing over 5 million images belonging to 80,000 people of varied nationalities and races (Guo, 2016).
Phase 2: Training:
With the Python API of TensorFlow (the most popular Deep Learning framework today), we implemented Python scripts to train the mask detection and face recognition models, and then saved the fit models to respective files in binary format.
Phase 3: Deployment:
With the C ++ API of TensorFlow, we ran the fit models to perform calculations and predications, and presented the predicted results of mask wearing and personal identity in the frontend interface.
For AI face recognition and mask wearing detection functions, we list the implementation details and technology sharing for developers' reference. For details, please visit Implementation Details for Developers.
Considering the difference in programming languages and operating systems between FLIR C3 SDK and users, we have customized a model implementation roadmap: model training by Python, model deployment by C++, and DLL compilation. With these steps, we have realized the productization of the project and achieved the given goals. This success will also provide valuable experiences for future AI programming and deployment practice in different operating environments.
In addition to AI, our Data Science and Innovation Team also leverages various innovative technologies, including blockchain, edge computing, augmented reality (AR) and virtual reality (VR), to enrich our whole range of innovative communications products, security solutions and managed services. With global knowledge, regional coverage and local services, we are the trusted Global Local ICT Solutions partner, providing customers with one-stop comprehensive ICT solutions. If you would like to know more about our products and services, please contact us.
You are about to visit our website