研究目的
To propose a three-dimensional (3-D)-printable frame and an open-source system to fabricate a wearable GT system with low-cost configuration and reasonable performance, and to develop the automatic discrimination of interest objects using the proposed GT and machine learning.
研究成果
The study successfully proposed a 3-D-printable frame for a GT system that is adjustable, wearable, and cost-effective. The GT system demonstrated reasonable performance with a 24-Hz sampling rate and an average accuracy of 2.58°. The application of the GT system for automatic discrimination of interest objects showed promising results with a 7-Hz sampling rate and a 3.8% classification error. The combination of GT and machine learning provides a foundation for further research in human attention analysis and interactive applications.
研究不足
The GT system's accuracy decreases at close distances, and the sampling rate of the application system is slower than the original GT system. The study also mentions the difficulty of automatically determining SOINN parameters for different input data.
1：Experimental Design and Method Selection:
The study focuses on designing a 3-D-printable frame for a GT system that is adjustable and wearable, using open-source design and software. The methodology includes the use of a 3-D printer for frame fabrication, two cameras for eye and front image recording, and infrared light sources for pupil detection.
2：Sample Selection and Data Sources:
The system is tested with a model of a 26-year-old woman, and the performance is evaluated based on the accuracy of gaze coordinates to a reference point.
3：List of Experimental Equipment and Materials:
Components include a Fosman USB camera, Microsoft Lifecam HD-6000, infrared LEDs, and PLA material for 3-D printing.
4：Experimental Procedures and Operational Workflow:
The process involves calibrating the GT system, tracking gaze coordinates, and applying machine learning for object discrimination.
5：Data Analysis Methods:
The accuracy of the GT system is calculated using linear regression and the coefficient of determination (Rsquare). The application system's performance is evaluated based on sampling rate and classification error.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
