AIRA features CPU-powered face recognition to address market challenges

News highlights

AIRA project manager Chia-Hsuan Hsieh

The AI revolution is reshaping the world, impacting every industry. AI being rapidly put to use has spurred surging demand for critical components. In particular, GPU being in constant shortage has many supply chain ramifications. Taiwan-based startup AIRA has developed a solution that stands out from its peers on the market to circumvent the GPU shortage. It uses Intel CPUs instead of GPUs to perform precise and instant face recognition, which allowed AIRA to receive the Intel Gold Partner status within just nine months after the company was founded, making it a rising AI star.

According to AIRA project manager Chia-Hsuan Hsieh, endeavoring smart image recognition technologies, AIRA introduced airaTrack, a face-tracking solution that captures the entire field in 2022. Its unique security implementations provide intelligent and complete protection for residential buildings, commercial complexes, offices, construction sites, and factories. Distinguished from the market, airaTrack uses Intel CPUs to carry out face recognition with no need for GPUs with ever-increasing prices, enabling significant savings in what it costs to have a system in place.

Hsieh added that AI-based face recognition solutions running on GPUs have a slew of challenges to overcome, the first being the cost. A 32-channel video surveillance server comes with a price tag of at least NT$300,000. Including routine maintenance fees, the operating expenses add up. Secondly, the system architecture is complicated. A face recognition system has to integrate multiple software development kits (SDK) to incorporate the functions it provides, requiring considerable time and effort for system development and integration. The third challenge lies in the acceptance check. For a face recognition system to deliver the specified accuracy, the software and hardware must synergize and exert their capabilities to the fullest. A system newly coming online can hardly perform up to standard and pass acceptance checks.

Running on Intel CPUs, airaTrack addresses all the above challenges. The SDK-based airaTrack enables significant reductions in the cost and time needed to get the system online, cutting the system setup cost to as low as NT$100,000. Moreover, the lower CPU operating temperature compared to GPUs allows airaTrack to adopt a fanless design, giving rise to higher flexibility in server space utilization and ease of maintenance. More importantly, as an integrated system combining completed software and hardware designs, airaTrack delivers the specified accuracy as soon as it comes online.

For its near-perfect accuracy and powerful capabilities, airaTrack is being used in wide-ranging application scenarios. The solution performs face recognition using more than a thousand feature points. Unless the subject has undergone major facial plastic surgeries, airaTrack can deliver high-accuracy face recognition regardless of appearance changes such as makeup and age.

Another key feature of airaTrack is continuous face tracking. For example, AIRA's solution targeting factory and retail applications can track faces across cameras. It can identify each face captured in a video frame and archive it as an index. For any selected target, it can search the face archive and find a match in an instant. In the example Hsieh presented, it took airaTrack only about ten minutes to search video footage totaling two to three hours captured by 16 cameras. This can be used to find a missing child in a retail setting and detect worker movements on a factory floor. Not only can the face recognition solution prevent trespassing and ensure factory security, but it can also replace badge access control and enable all kinds of smart employee benefits applications.

AIRA also has a solution to address privacy concerns around face recognition. There is no need to import face data before face tracking. The face feature values generated during face tracking are encrypted and protected against reverse engineering. Only AIRA's systems can decode and do crosschecks, thereby guaranteeing privacy.

For its high level of privacy protection, system setup flexibility, and system integration, on top of friendly cost, airaTrack is enthusiastically embraced by the market. AIRA is working with multiple international companies and among them, Amazon Web Services (AWS) is a key partner that provides complete technical support in cloud-based AI. Joining forces with AWS, AIRA has developed an AI and IoT integrated solution, to target factory security and smart manufacturing applications going forward. The partnership will enable AIRA to accelerate time-to-market and safeguard enterprise security.

AIRA's airaTrack, a full-field face tracking solution

AIRA's airaTrack, a full-field face-tracking solution
Photo: AIRA