Friday 07 March 2025
The quest for a more empathetic AI has led researchers to develop a facial emotion detection interface integrated into a mobile humanoid robot, capable of displaying real-time emotions from multiple individuals on a user-friendly graphical interface.
At its core, the system relies on various deep neural network models fine-tuned on the FER2013 dataset, with each model optimized using Adam with a learning rate of 0.0001 and strategies like EarlyStopping and ReduceLROnPlateau to prevent overfitting and dynamically adjust the learning rate. The accuracy and memory footprint of each model are carefully considered to ensure compatibility with both personal computers and mobile robots.
The system’s performance is put to the test in two sets of experiments, where a single participant interacts with the robot displaying a range of emotions, followed by multiple participants engaging in various interactions simultaneously. The results demonstrate the system’s ability to accurately detect faces and classify emotional states in real-time, even when faced with multiple individuals.
One of the key challenges in developing this system is the need for a balance between model size and accuracy. Larger models may achieve higher accuracy rates but come at the cost of increased memory requirements, making them less suitable for deployment on resource-constrained devices like mobile robots.
The researchers’ approach addresses this challenge by selecting a model that strikes a balance between accuracy and memory footprint. The EfficientNetV2- B0 model, with an accuracy rate of 70.00% and a memory footprint of 139 MB, is chosen for implementation on the robot due to its compact size and high performance.
The system’s user-friendly interface allows users to capture live video streams using their device’s camera, which are then processed in real-time to detect and classify facial expressions. For single-face emotion recognition, the application highlights the detected face and displays the identified emotion with corresponding confidence levels. In multi-face scenarios, the interface efficiently detects multiple faces within the same frame, assigning emotions to each detected face individually.
The results of this research have significant implications for human-robot interaction and affective computing. The ability to detect and classify emotional states in real-time can enable robots to better understand and respond to human needs, potentially leading to more effective and empathetic interactions.
In addition to its practical applications, the system’s development also highlights the importance of considering factors like model size and accuracy when designing AI systems for deployment on resource-constrained devices.
Cite this article: “Facial Emotion Detection Interface for Human-Robot Interaction”, The Science Archive, 2025.
Artificial Intelligence, Facial Emotion Detection, Mobile Humanoid Robot, Deep Neural Networks, Fer2013 Dataset, Emotional State Classification, Real-Time Processing, Human-Robot Interaction, Affective Computing, Model Size And Accuracy.
