top of page

Deploying a Facial Recognition Access Control System with Thermal Monitoring

New to Facial Recognition Technology? How do I deploy a Facial Recognition Access Control with Thermal Scanner solutions in COVID-19 situation?

Here is Facial Recognition 101 and Thermal Scanner deployment tips:

With Facial Recognition Access Control with Temperature measurement solutions available in the Singapore market, the interest, use and fear of Facial Recognition needs to be addressed. With this article, it will shed some light on Facial Recognition technology fundamentals and Pensees’ R&D capabilities. 1. Facial Detection Facial Recognition performance is affected by 3 core factors: - Angle of Faces: Access Control system will usually require user to look directly at the camera, performance can be affected how a person tilts their head (down, left, right, etc.). - Lighting of Faces: Performance will be affected significantly depending on the lighting conditions of the environment, like shadows, darkness, complicated backgrounds, backlight. - Computing load to detect faces: Finding and determining a Face amongst the environment and sub-standard Access Control devices have limited processing power.

Pensees’ elegant product design includes users’ consideration and assists companies in their deployment of Access Control systems. Illumination strip on the top of the device will be enabled when detecting the presence of a user, this gives added lighting to harsher conditions.

Companies can change the screensaver of the device to information that they would convey to user or to display beautiful designs to capture user’s attention to the device.

Pensees deploys Facial Recognition algorithm that won the Special Innovation Award and that emerged first ranking in 3 competitions track for International Conference on Computer Vision (ICCV) 2019 Light Weight Face Recognition Challenge.

In light of face recognition applications with large-scale database, it is still a challenge to find a model of lightweight and high precision that has excellent performance under unconstrained dynamic scenes of video and image surveillance. Lightweight Face Recognition Challenge (LFRC) was set up for the mentioned purpose. The competition became an important one at ICCV 2019, attracting 292 teams from all over the world. Unlike other face recognition competitions, ICCV LFRC posed strict limits on complexity of your model and training data, and Pensees overcame the limitations and emerge top ranking, thus displaying its computational optimization capabilities in limited processing power devices like Access Control systems. 2. Lighting Issues Facial recognition requires detailed facial details and image clarity. Many cameras include IR LEDs to illuminate dark areas and increase details captured by the camera. However, IR performance varies significantly which can decrease details and decrease face recognition confidence.

Pensees’ Access Control System are all equipped with Infrared cameras with Liveness Detection capability. Pensees’ technical team takes pride in our training and Artificial Intelligence- Facial Recognition capabilities, our professional technical support will advise customers on their deployment location and environment to steer clear of performance issues. 3. Accuracy rate and False Alarm rate Measuring accuracy of facial recognition systems is not easy. Many vendors tend to market with simplistic, impressive-sounding metrics like 99.99% accurate. However, AI scientists measure facial recognition much differently, with tradeoffs between false positive (i.e., matching the wrong face to someone else) and false negatives (failing to match a face to a person on the watchlist).

NIST is a U.S Government standards and technology department that tests face recognition algorithms as part of their ongoing Face Recognition Vendor Testing (FRVT). Strong NIST results are cited by many manufacturers as a marketing tool, however, it can be difficult for non-experts to understand the details of the results, which are abstract and academic.

Pensees’ world record holder in 2019 for NIST IJB-B and IJB-C and in the ICCV LRFC challenge, we achieved the highest accuracy/the lowest false negative rate under different false positive requirements (for example under 1e-5, means under the case there is only one false positive happens among 100,000 faces search).

4. Liveness Detection/Anti-spoofing: Printed Photo/Video confidence recognition Access Control with Facial Recognition will require Liveness Detection to ensure the security of your premises and for companies who are deploying the system for attendance taking, Liveness Detection becomes a crucial requirement.

Facial Liveness detection can be achieved via a number of heuristic methods, with three of the more pragmatic examples being: Mouth/Lip Movement: For liveness confirmation, users moving their mouth to breathe or speak 'proves' they are not a printed image. Eye Movement: In the same way, human eyes are not stationary, and even subtle movements of eyes indicate subjects are live people and not images. Blink Detection: Photo images do not blink, so visual confirmation faces/eyes are blinking is a common liveness test.

Like other aspects of computer vision, actual result varies depending on image quality and which algorithm is deployed. Facial Recognition that includes Liveness Detection is an advanced algorithm, many companies commonly do not offer it and can be expensive to include as it might require even specialized hardware.

Pensees’ has implemented solutions over 50 cities in China and consolidated advanced capabilities in ensuring the Liveness Detection technology in their products.

5. Enforcement of wearing mask, and Face recognition with masks Due to Covid-19 wearing mask becomes compulsory even now in Singapore. Mask will become common in public places and wore not only just in medical facilities, due to the Covid-19 situation. Medical masks significantly decrease match confidence, and with most people generally not recognized at all.

Pensees’ in-housed developed Deep Learning based patented Face recognition with mask algorithm and mask wearing detection are locally developed here in Singapore by our Pensees R&D Center led by Ms Jane Shen who has over 20 years of AI - Computer Vision experience. She and her team had developed many technologies have been deployed in many products and applications. Pensees’ product can detect if you are wearing or not wearing a mask and will remind you to wear a mask.

6. Localisation of Facial Recognition capabilities For successful deployment of Face Recognition systems in Singapore, it is fundamental to check if the system has been deployed in our multi-racial and diversified community. Always make sure if the R&D team has designed and trained its good algorithm hence reducing possible biasness to results, causing accuracy issues and high false positive rates which is unacceptable for Access Control system to be used as a security device or attendance taking.

Pensees’ R&D center in Singapore and Beijing, China, Ms Jane Shen, Chief Scientist of Pensees Group and her main team of AI Scientist members are based here in Singapore. The team supports the deployment and calibration of all our products in Singapore and the region and is familiar with the limitations of facial recognition algorithms in diversified settings. The team overcame limitations with their R&D efforts and would continue to strive to keep the state-of-art standards. Pensees’ technical team works with customers closely to understand, calibrate and make adjustments for Facial Recognition solutions to be successfully installed which is fundamental to technology deployment.

7. Thermal Camera on Access Control deployment The following are fundamentals of thermal camera deployment: - Subject distance: The subject must be at of a stated distance range for screening. - Ambient temperature issues: How and where the thermal camera is placed will affect the accuracy of temperature reading, if the device is in direct sunlight, reflective light, heated area (near car engines), significant estimated temperature swings, reducing or increased estimated temperature will be read by the device. It is impossible for thermal cameras to claim to be weather-proof or ambient temperature proof. - Face detection issues: Attempted measurements when subjects were looking down or away from the camera, resulting in incorrect measurements as the hottest parts of the face were not visible.

In Singapore, when an individual is out in the direct sunlight for a period of time or had rushed to the location and attempted to have their temperature read by thermal devices, their temperature might momentarily increase. However, if the individual stood still and cools down even for a minute, the body temperature will be reduced to a normal range. This similarly happens with other thermal guns and medical thermometer.

Pensees’ Technical Installation team will assist customers in deploying solution in the most ideal location and is upfront about limitations of thermal cameras deployment as mentioned above.

About Pensees Pensees is a company with an international leading AIoT ecosystem platform. It takes “AI is a service” as its mission and transforms cutting-edge AI technologies into inclusive intelligent service evolution.

Based on its world’s leading full-scenarios intelligent visual processing brain, the company has been building an AIoT ecosystem platform that has full-stack computer vision technologies and continues to lead innovations in intelligent edge devices. Since its establishment, the AI research team of the company has held 14 world-records in international AI competitions.

A closed-loop model of AI, IoT, and SaaS has been leveraged by the company to provide AIoT devices, cloud services, and scenario-based AIoT solutions in fields of smart cities, intelligent residential communities, intelligent businesses, and more, so as to promote the industrialization of AI.



bottom of page