Object classification through heterogeneous fog with a fast data-driven algorithm using a low-cost single-photon...

Area of Science:

• Computer Vision
• Sensor Technology
• Machine Learning

Background:

• Pose classification is challenging in adverse weather conditions like fog.
• Existing methods may struggle with dynamic and heterogeneous fog environments.
• Low-cost sensors are desirable for widespread consumer electronics applications.

Purpose of the Study:

• To develop and evaluate a framework for classifying wooden mannequin poses using a single-photon avalanche diode (SPAD) array.
• To assess the performance of an extreme learning machine (ELM) for rapid pose classification in fog.
• To demonstrate the feasibility of real-time pose classification with a low-cost sensor system.

Main Methods:

• Utilized a low-cost SPAD array sensor with a diffused VCSEL laser within a fog chamber.
• Collected training datasets by configuring sensor firmware for temporal and spatial resolutions.
• Trained an extreme learning machine (ELM) for pose classification, benchmarking against Convolutional Neural Networks (CNN).

Main Results:

• The ELM achieved 90.65% accuracy in training and 89.58% in testing for pose classification.
• Accurate classification was demonstrated up to 35 cm in dynamic heavy fog.
• Both ELM and CNN showed robustness as fog density increased.

Conclusions:

• The proposed ELM-based framework effectively classifies mannequin poses in dynamic, heterogeneous fog using a low-cost SPAD sensor.
• The system enables real-time pose classification applications in consumer electronics.
• The study highlights sensor limitations and future research directions for improved optical sensing.