Driver's Facial Expression Recognition in Real-Time for Safe Driving

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep neural networks (DNNs) offer advanced facial expression recognition (FER) but demand substantial computational resources, limiting their use in hardware-constrained environments.
• Conventional machine learning methods for FER have limitations that DNNs aim to overcome, yet the high cost of DNNs presents a new challenge for practical applications.

Purpose of the Study:

• To develop a fast and efficient facial expression recognition (FER) algorithm suitable for low-specification automotive devices.
• To enable real-time driver emotion monitoring without the high memory and processing costs associated with deep learning approaches.

Main Methods:

• A hierarchical weighted random forest (WRF) classifier was developed, trained on data similarity to enhance accuracy.
• Facial landmarks are detected, and geometric features are extracted based on landmark spatial positions.
• The extracted features are fed into the hierarchical WRF classifier for facial expression classification.

Main Results:

• The proposed method achieved high accuracy, comparable to deep learning FER methods (92.6% on CK+, 76.7% on MMI).
• Processing cost was significantly reduced, approximately 3731 times less than deep learning methods.
• Experimental validation was conducted using the CK+, MMI, and KMU-FED databases.

Conclusions:

• The proposed hierarchical WRF-based FER algorithm is highly efficient and accurate.
• This method is optimized for real-time embedded applications with limited computing resources, such as in-vehicle systems.
• The approach offers a practical alternative to computationally expensive DNNs for driver emotion monitoring.