Efficient automated high dynamic range 3D measurement via deep reinforcement learning

• DOI: 10.1364/oe.510515
• Publication Date: 2024-01-08T10:00:09Z
• Authors: Pan Zhang; Kai Zhong; Zhongwei Li; Yusheng Shi
• Keywords: 0103 physical sciences; 01 natural sciences

High dynamic range 3D measurement technology, utilizing multiple exposures, is pivotal in industrial metrology. However, selecting the optimal exposure sequence to balance efficiency and quality remains challenging. This study reinterprets this challenge as a Markov decision problem presents an innovative selection method rooted deep reinforcement learning. Our approach’s foundation image prediction network (EIPN), designed predict images under specific thereby simulating virtual environment. Concurrently, we establish reward function that amalgamates considerations of number, time, coverage, accuracy, providing comprehensive task definition precise feedback. Building upon these foundational elements, (ESN) emerges centerpiece our strategy, acting decisively agent derive selection. Experiments prove proposed can obtain similar coverage (0.997 vs. 1) precision (0.0263 mm 0.0230 mm) with fewer exposures (generally 4) compared results 20 exposures.