A comparison of Arabic sign language dynamic gesture recognition models

Area of Science:

• Computer Science
• Human-Computer Interaction
• Linguistics

Background:

• Arabic Sign Language (ArSL) is a vital communication method for the deaf community.
• Limited research exists on sensor-based ArSL recognition and natural user interfaces.
• No single classifier approach is universally optimal for hand gesture recognition.

Purpose of the Study:

• To investigate optimal algorithm and parameter combinations for accurate ArSL gesture recognition.
• To propose and evaluate a dynamic prototype model (DPM) for ArSL recognition.
• To enhance accessibility and communication for ArSL users through technology.

Main Methods:

• Developed a dynamic prototype model (DPM) utilizing the Kinect sensor.
• Implemented and tested eleven predictive models based on Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN) algorithms.
• Varied parameter settings for each algorithm to assess their impact on recognition accuracy.

Main Results:

• Support Vector Machine (SVM) models demonstrated superior performance in recognizing dynamic ArSL words.
• Optimal accuracy was achieved with SVM models employing a linear kernel and a cost parameter of 0.035.
• The DPM effectively recognized specific dynamic ArSL gestures.

Conclusions:

• The combination of SVM with a linear kernel and a cost parameter of 0.035 is highly effective for ArSL gesture recognition.
• Sensor-based systems, particularly with optimized SVM models, show significant potential for ArSL interpretation.
• Further research can build upon these findings to develop more sophisticated ArSL recognition systems.