We have developed a highly accurate and robust AI model capable of estimating both age and gender even when the person is wearing a face mask. Unlike conventional models that struggle with occlusions, our system is trained to recognize subtle facial features, contours, and patterns that remain visible even when key areas like the mouth and nose are covered.
This makes it particularly effective in real-world scenarios where mask-wearing is common, such as in healthcare settings, public transportation, and retail environments. With its ability to maintain high prediction accuracy despite facial obstruction, this model represents a significant advancement in AI-based facial analysis.
In response to the challenges posed by the COVID-19 pandemic, particularly the difficulty in accurately estimating age and gender due to mandatory mask-wearing in Japan, the client has initiated this project. The goal of this project is to develop an Edge AI application that accurately determines the age and gender of individuals captured by surveillance cameras. Unlike the previous approach that was used by the client, where age and gender estimation models were developed separately, our added value lies in integrating these models to estimate both the age and gender simultaneously with an improved accuracy.
The project faced several challenges. One major issue was **age determination models exhibiting gender-based inaccuracies**, which particularly affected women, leading to biased or unreliable results. Additionally, the models had **lower accuracy when dealing with masked faces**, which impacted the overall reliability of age estimation in real-world scenarios where people are often wearing masks.
Another challenge was the **compatibility issues with a variety of models**, which hindered performance, especially on low-spec devices like the **Jetson Nano**. This made it difficult to ensure smooth and efficient operation across different hardware, limiting the flexibility and scalability of the solution.
To address the challenges faced in age and gender estimation, several solutions were implemented to enhance the system’s accuracy and efficiency. The existing model was modified to enable simultaneous **age and gender estimation** within a single framework, reducing the need for separate models and improving overall system efficiency. To tackle the issue of **masked faces**, **transfer learning strategies** were adopted, with the model trained separately on **masked and unmasked face datasets**. This approach helped the model accurately discern facial features, even in the presence of masks. Additionally, to ensure compatibility with **low-spec devices** like the **Jetson Nano**, **lightweight models** were developed and optimized, allowing for seamless operation without compromising performance or accuracy, making the solution accessible to a broader range of hardware platforms.
We are excited to introduce our advanced AI-based age and gender detection model, which can now simultaneously estimate both age and gender using a single unified architecture. This integrated approach marks a significant improvement over traditional models that rely on separate systems for each task. By combining the two into one streamlined framework, we’ve achieved faster processing, reduced computational load, and improved overall accuracy, making it ideal for real-time applications across various industries.
Built using cutting-edge deep learning techniques and trained on a diverse dataset, our model demonstrates higher performance and reliability even under challenging conditions such as varied lighting, facial angles, and backgrounds. This innovation opens new possibilities for applications in retail analytics, surveillance, healthcare, and more—delivering accurate, efficient, and scalable results that set a new standard for age and gender detection in AI.
Our mission is to empower organizations to embrace innovation, overcome challenges, and drive growth.
Featured Article
