rPPG Model Licensing and Deployment: Options Explained

The integration of remote photoplethysmography (rPPG) into hardware is accelerating, moving from a niche technology to a key feature for hardware OEMs, automotive suppliers, and IoT device makers. As the demand for contactless health monitoring grows, so does the complexity of bringing rPPG to market. The global Photoplethysmography (PPG) biosensors market was valued at over $450 million in 2023 and is projected to grow at a CAGR of over 10%, making the strategic decisions around rPPG model licensing and deployment more critical than ever. Choosing the right path determines not just the initial cost, but the long-term performance, scalability, and defensibility of the product.

"The global PPG biosensors market size was estimated at USD 459 million in 2023 and is expected to grow at a compound annual growth rate (CAGR) of 10.8% from 2024 to 2030." - Grand View Research, 2023.

Navigating rPPG model licensing and deployment options

For hardware teams, the central challenge is deciding how to acquire and implement an rPPG model that works reliably with their specific camera and hardware constraints. There is no one-size-fits-all answer, and the optimal choice depends on factors like in-house expertise, budget, and the uniqueness of the operating environment. The main rPPG model licensing deployment options fall into three broad categories: licensing a commercial SDK, building from open-source, or commissioning a custom-trained model.

Each approach has significant trade-offs in terms of upfront cost, time to market, and final model accuracy. A commercial SDK may offer the fastest path to a proof-of-concept, but may lack the specificity required for a novel camera sensor or challenging lighting conditions. Open-source models provide transparency and control, but demand significant in-house data science and machine learning expertise to adapt and validate. A custom build offers the highest potential for performance, but requires a close partnership with a specialized vendor.

Licensing Model	Key Characteristics	Best For
Commercial SDK	Pre-trained, general-purpose models. Fast integration via API/SDK.	Teams needing to quickly validate an idea or with standard camera hardware.
Open-Source	Publicly available code and models. Requires significant in-house expertise.	Academic research or teams with deep AI talent and unique requirements.
Custom Build	Model trained specifically for the OEM's camera and use case.	High-performance applications and unique hardware (e.g., IR cameras, automotive).

Deployment strategies: from cloud to edge

Once a model is licensed or built, the next decision is deployment. This is not a trivial step; the deployment architecture has major implications for user experience, privacy, and operating cost.

• Cloud-based API: The simplest deployment model. Video frames are sent to a cloud server, which processes the data and returns the vital sign results. This approach is easy to implement but introduces latency and privacy concerns.
• On-Device/Edge Deployment: The rPPG model runs directly on the device's processor (e.g., an IoT device's SoC or a car's computer). This minimizes latency and keeps all data local, which is critical for privacy-sensitive applications. However, it requires a highly optimized model that can run efficiently on resource-constrained hardware.
• Hybrid Model: A combination of edge and cloud. The device might perform initial signal processing on the edge to reduce the amount of data sent to the cloud, with more intensive calculations happening on the server.

Industry Applications

The choice of licensing and deployment is heavily influenced by the target application.

Automotive driver monitoring

In-cabin sensing is a primary use case for rPPG. Automotive OEMs and Tier-1 suppliers are integrating driver monitoring systems to meet safety regulations and enhance the driver experience. These systems often use infrared (IR) cameras to ensure performance in all lighting conditions. The unique camera hardware and the safety-critical nature of the application make custom-trained, edge-deployed models the most common choice.

Iot and smart home devices

For smart home hubs, mirrors, and other connected devices, rPPG can provide ambient wellness monitoring. The hardware is often cost-sensitive, and the use case may be less demanding than clinical or automotive applications. Here, a commercial SDK with an edge deployment option can provide a good balance of performance and cost.

Clinical and telehealth kiosks

Kiosks in pharmacies or clinics require high accuracy and reliability. While the camera and lighting are more controlled than in other environments, the need for robust performance across different populations is critical. These applications often start with a commercial SDK and then move to a custom-tuned model to improve accuracy for their specific user base.

Current research and evidence

The field of rPPG is evolving rapidly, with researchers continually improving the accuracy and robustness of the algorithms. Much of the foundational research has been conducted by academic institutions and published in peer-reviewed journals. For example, work by W. Wang and colleagues at the Chinese Academy of Sciences (2020) has been influential in developing deep learning architectures for rPPG. Many open-source toolkits, such as the one maintained by the University of Cambridge, provide a baseline for new research and commercial products. The primary focus of current research is on improving performance in challenging real-world conditions, such as low light, subject motion, and diverse skin tones.

The future of rPPG model deployment

The future of rPPG is on the edge. As machine learning models become more efficient and embedded processors become more powerful, the default deployment strategy will be on-device. This trend is driven by several factors:

• Privacy: Keeping health data on the device is a major selling point for consumers and a requirement in many regulated industries.
• Latency: Real-time feedback is critical for many rPPG applications, from driver drowsiness alerts to biofeedback for wellness apps.
• Cost: Processing video streams in the cloud can be expensive at scale. Edge deployment eliminates this ongoing cost.

We are also likely to see more hybrid licensing models, where companies start with a commercial SDK for prototyping and then work with the vendor to create a custom-tuned model for their production hardware.

Frequently asked questions

What is the most common rPPG licensing model? For companies new to rPPG, licensing a commercial SDK is the most common starting point due to its speed and relative ease of integration. However, as products mature, many companies move to custom models to achieve higher performance.

Can I use an open-source rPPG model in a commercial product? Yes, but it requires careful attention to the license terms. Many open-source projects are available under permissive licenses like MIT, but they may not include the rights to use pre-trained models for commercial purposes. Furthermore, the OEM is responsible for all validation and regulatory compliance.

How does camera choice affect deployment? The camera is the most critical hardware component for rPPG. A model trained on data from one camera will not perform optimally on another. This is why camera-specific model training and tuning are essential for high-performance applications. This is especially true for non-standard cameras, such as the IR and thermal cameras used in automotive and industrial settings.

As hardware teams work through the landscape of rPPG model licensing and deployment options, it is clear that the path to a successful product requires a strategic approach. The decision is not just about technology, but about finding the right partner to de-risk the integration process and ensure the final product meets the high standards of the market. Circadify specializes in creating custom-trained rPPG models for unique hardware and challenging use cases. To learn more about a custom build for your device, contact our experts at circadify.com/custom-builds.