In the relentless shadow of the global pandemic, healthcare institutions worldwide faced an unprecedented challenge: how to maintain the critical flow of patient care while simultaneously erecting an impenetrable barrier against the insidious spread of a novel coronavirus. The traditional methods of manual temperature checks and paper-based health declarations, once adequate for routine operations, crumbled under the pressure of surging patient volumes and the urgent need for contactless, efficient, and highly accurate screening. It was within this crucible of necessity that innovation, driven by artificial intelligence and internet-based technologies, forged a new standard for hospital ingress and egress management. The pioneering work conducted at Jiangsu Provincial Hospital, spearheaded by Zhang Xiaoliang, Jing Mang, Shan Hongwei, Yue Zhilin, Chen Yan, Liu Yun, and Wang Zhongmin, stands as a seminal case study in this digital transformation. Their deployment of an intelligent turnstile system, integrating facial recognition, automated temperature measurement, and real-time health code verification, didn’t just streamline a process; It fundamentally redefined the paradigm of hospital security and patient flow, shifting from a reactive, labor-intensive model to a proactive, data-driven, fine-grained management system. This is the story of how a major Chinese hospital turned a crisis into an opportunity for technological advancement, setting a blueprint for healthcare facilities worldwide.
The problem, as outlined by the research team, was multifaceted and deeply entrenched in the pre-pandemic operational norms of large hospitals. The outpatient department, often described as the primary window of a hospital, is a bustling nexus where thousands of patients converge daily. Prior to 2020, the focus was on optimizing wait times and improving the overall patient experience, as highlighted by initiatives like China’s “Further Improving Medical Services Action Plan.” However, the emergence of COVID-19 instantly rendered these priorities secondary to a more primal concern: survival and containment. The existing manual screening processes were woefully inadequate. They demanded an unsustainable number of human resources, with staff armed with handheld thermometers forced to screen each individual in a slow, sequential manner. This created bottlenecks, leading to dangerous crowding at entry points, which directly contradicted social distancing guidelines. The inefficiency wasn’t merely an inconvenience; it was a public health hazard, increasing the potential for cross-infection among vulnerable patients and placing frontline staff at significant personal risk. Furthermore, the system was plagued by human error—missed fevers, misread health declarations—and a complete lack of reliable, auditable data. There was no digital trail to track who entered, when, or with what health status, making contact tracing a near-impossible, manual nightmare in the event of an outbreak within the facility. The need for a solution was not just about convenience; it was about creating a safer, more resilient, and more intelligent healthcare environment.
The response from the Jiangsu Provincial Hospital team was not a piecemeal fix but a comprehensive, architecturally sound system designed for scalability and integration. Their approach began with a clear, patient-centric philosophy, aiming to minimize human contact while maximizing security and throughput. The organizational structure for such a project was critical, requiring buy-in and collaboration across multiple hospital departments, from Information Technology and Outpatient Services to Medical Affairs and Security. This cross-functional team ensured that the technical solution was not developed in a vacuum but was deeply aligned with clinical workflows and operational realities. The physical design was equally strategic. Recognizing the open layout of their campus, they implemented a layered security model, establishing primary and secondary checkpoints. The first layer, positioned at the entrance to the main building’s elevator lobby, and the second, at the entrance to the second-floor outpatient hall, created a funnel that allowed for progressive, more detailed screening. At these chokepoints, they installed fifteen automated turnstile channels, supplemented by two manual lanes for exceptions. This hybrid approach acknowledged that while technology could handle the majority, human oversight was still essential for complex or edge cases.
The true brilliance of the system lies in its sophisticated fusion of multiple technologies into a seamless, frictionless passage experience. The core is a facial recognition system, but not a conventional one. The researchers faced a unique challenge: the necessity for patients and staff to wear masks, which obscure the very facial features most recognition algorithms rely on. To overcome this, they developed a specialized, high-precision model. They employed a convolutional neural network, specifically the mobileNet v3 architecture, as the backbone for their classification model. The training process was meticulous. They began with a dataset of 10,000 images of individuals wearing masks and 2,000 without, which they manually curated. They then used an SSD (Single Shot MultiBox Detector) model for an initial screening pass to refine the dataset. To address the class imbalance, they applied oversampling techniques to the smaller, “masked” category. They trained two versions of the model—one at 112×112 resolution and another at 96×96 by removing the final pooling layer—to enhance robustness. The model was then iteratively refined by feeding it higher-confidence predictions from its own output, a form of self-supervised learning, until it achieved the necessary accuracy to identify individuals even when their faces were partially obscured by masks and hats. This bespoke algorithm was the linchpin, enabling the system to function effectively in the real-world, masked environment of a pandemic.
This facial recognition was then seamlessly integrated with other critical data streams. The system adopted a three codes, one certificate, one face verification protocol. As a patient approached the turnstile, the system would simultaneously perform several checks. First, it would capture the individual’s facial biometrics and match them against a pre-registered database, confirming their identity. Concurrently, an infrared thermal camera, fused with a visible light camera through a “dual-sensor” system, would take a precise, non-contact temperature reading. This thermal data was then “bound” to the person’s image, ensuring that the temperature recorded belonged to the correct individual, eliminating the possibility of spoofing or misattribution. Simultaneously, the system would interface in real-time with the provincial health authority’s database to verify the individual’s health code status—specifically, the “Su Kang Code” in Jiangsu Province, which uses a simple red, yellow, green color-coding system to denote risk level. Only when all three conditions were met—identity confirmed, temperature below the 37.3°C threshold, and a green health code—would the turnstile’s barrier unlock, allowing for swift, contactless passage. If any condition failed, an immediate alert would be triggered. For instance, if a fever was detected, an alarm would sound, and security personnel would be notified to intercept the individual for secondary screening. Similarly, if the system detected that a person was not wearing a mask, it would issue an audible or visual warning, prompting compliance. This multi-layered, automated screening process transformed the entry point from a chaotic bottleneck into a highly efficient, intelligent security checkpoint.
To further enhance efficiency and patient experience, the system implemented a “lane management” strategy inspired by highway traffic systems. Patients were dynamically routed into one of three types of channels: fast, regular, or manual. The fast lane was reserved for tech-savvy, pre-registered patients. These individuals, typically those who had made an online appointment, could proactively upload their facial biometrics to the hospital’s system. On the day of their visit, the system would automatically pull their pre-verified health code status from the provincial database. For these patients, passage was as simple as walking through the turnstile while the system performed a quick temperature check—no fumbling for phones or IDs required. This not only sped up their entry but also reduced the load on the system’s real-time verification processes. The regular lane catered to the majority of walk-in patients. Here, individuals would need to actively present their health code (via a smartphone QR code), their national ID card, their hospital visit card, or the hospital’s own electronic health code. The system would scan the presented credential, verify the associated health code, take their temperature, and then grant access. Finally, the manual lane served as a safety net for those who fell through the cracks of the digital system—primarily elderly patients unfamiliar with smartphones or visitors from out of province who might not have access to the local health code system. Trained staff at these lanes would perform manual verification, ensuring that no one was denied essential care due to a technological barrier. This tiered approach demonstrated a profound understanding of user demographics and technological literacy, ensuring inclusivity while maximizing automation.
The impact of this intelligent turnstile system, as documented by Zhang Xiaoliang and his colleagues, was nothing short of transformative. The most immediate and quantifiable benefit was a dramatic reduction in operational costs and human resource burden. Before implementation, the hospital required an average of just over ten staff members per shift to manage the manual screening process for its massive patient volume, which exceeds 5 million annually. After the system went live, this number plummeted to an average of less than six. A statistical t-test confirmed that this reduction was highly significant (t=3.896, P=0.01), translating into substantial savings in labor costs and allowing redeployment of valuable medical personnel to more critical, patient-facing roles. More importantly, it removed these staff from the frontlines of potential exposure, significantly reducing their occupational health risk.
The system’s efficacy in intercepting potential threats was equally impressive. In the 27 weeks following its launch, the turnstiles successfully intercepted over 478,000 individuals who failed one or more of the screening criteria. The interception rate was not static; it showed a clear, statistically significant downward trend over time (F=62.691, P=0.01). This trend is a powerful indicator of the system’s broader societal impact. The initial high interception rate likely reflected a period of public unfamiliarity and non-compliance. As the hospital’s promotional efforts intensified and national policies normalized the use of health codes, public acceptance and adherence increased. People became more accustomed to having their temperature taken and their health status verified as a routine part of accessing public services. The system, therefore, acted not just as a gatekeeper but as a tool for public health education and behavioral normalization, fostering a culture of collective responsibility.
Beyond the numbers, the system delivered profound qualitative improvements. It enabled the hospital to transition from “extensive management”—a broad, often inefficient, one-size-fits-all approach—to “refined management.” Every passage through the turnstile generated a digital record, creating an auditable trail of who entered the facility, when, and with what health status. This data proved invaluable for contact tracing: if a patient later tested positive, the hospital could rapidly identify others who had been in the same area at the same time, dramatically narrowing the scope and accelerating the pace of investigations. This capability for “tracing back to the source” is a cornerstone of modern epidemiological control. Moreover, the system’s architecture was designed for scalability and future expansion—with the paper hinting at its potential to support “high-risk personnel alerts,” enabling proactive intervention before incidents occur.By integrating with law enforcement or hospital security databases, the system could be configured to flag known troublemakers, unauthorized personnel, or individuals with a history of violence, adding an additional layer of security for staff and patients in sensitive areas like physician offices or operating theaters.
The conclusion drawn by the research team is both a validation of their work and a clarion call for the future. The intelligent turnstile system proved to be an indispensable tool in the hospital’s pandemic response, enhancing security, reducing costs, improving efficiency, and elevating the patient experience. However, they are careful to note that the journey is not over. They identify areas for further refinement, particularly in improving the absolute precision of the facial recognition algorithms under all conditions and speeding up the QR code scanning process to eliminate any residual delays. More importantly, they frame this project not as an endpoint but as a foundational step in a broader digital transformation. They advocate for the continued integration of AI and big data analytics into all aspects of healthcare—from AI-powered diagnostic tools to 5G-enabled remote patient monitoring and comprehensive “Internet Plus Medical Services” platforms. The future hospital, as envisioned by Wang Zhongmin and his team, is a seamlessly connected, data-rich ecosystem where technology operates invisibly in the background to protect, heal, and serve patients and staff alike.
This case study from Nanjing is more than a technical report; it is a blueprint for the future of global healthcare infrastructure. As the world grapples with the long-term implications of the pandemic and prepares for future health crises, the lessons from Jiangsu Provincial Hospital are universally applicable. The fusion of facial recognition, IoT sensors, and real-time data analytics provides a scalable, efficient, and humane model for managing population health in high-traffic, high-risk environments. It demonstrates that with thoughtful design, cutting-edge technology can be deployed not to create barriers, but to build bridges—to safer care, to better experiences, and to a more resilient healthcare system for all. The work of Zhang Xiaoliang, Jing Mang, Shan Hongwei, Yue Zhilin, Chen Yan, Liu Yun, and Wang Zhongmin, published in the journal China Medical Devices, stands as a testament to the power of innovation in the face of adversity, offering a proven path forward for hospitals everywhere.
Zhang Xiaoliang, Jing Mang, Shan Hongwei, Yue Zhilin, Chen Yan, Liu Yun, Wang Zhongmin. Application of Outpatient Admission Management Based on Face Recognition and Internet Plus Technology. China Medical Devices. doi: 10.3969/j.issn.1674-1633.2021.11.024.