Real-time localization and classification of the fast-moving target based on complementary single-pixel detection.

Area of Science:

• Optoelectronics
• Computational Imaging
• Pattern Recognition

Background:

• Traditional imaging struggles with real-time localization and classification of fast objects due to data needs, slow rates, and motion blur.
• Non-visible wavelength imaging presents additional challenges for dynamic object analysis.

Purpose of the Study:

• To develop an image-free method for simultaneous real-time target localization and classification.
• To integrate target localization and classification into a unified framework using complementary single-pixel detection.

Main Methods:

• Utilized complementary single-pixel detection with four specific illumination patterns.
• Employed centralized geometric moments for target localization and classification.
• Achieved simultaneous determination of centroid position and target shape.

Main Results:

• The proposed method achieved an update rate of up to 5.55 kHz.
• Experimental results showed root-mean-square error (RMSE) for centroid localization below 0.5 pixels.
• Achieved 93.3% classification accuracy for 30 different objects under diverse conditions.

Conclusions:

• The image-free method offers robust and accurate real-time localization and classification of fast-moving objects.
• Demonstrated strong adaptability in complicated environments, outperforming traditional imaging.
• Potential applications include target tracking, character recognition, industrial automation, and optoelectronic neural networks.