基于深度学习的3D激光测量图像反光区域分离方法

Separation Method of Reflective Area of 3D Laser Measurement Image Based on Deep Learning

提出了基于深度学习的3D激光测量图像中反光区域准确且稳定的分离方法。基于深度学习模型U-net网络实现了对3D激光测量图像中反光区域和激光线区域的语义分割;图像语义分割实现了对测量图像中不同区域的高精度分离。反光区域分离后的图像能够减少反光对中心线提取的干扰,研究结果表明,基于深度学习的激光测量图像区域分割可以更加精确地实现激光条纹的提取,同时保证提取结果的稳定性。单幅激光测量图像的区域分割时间仅为2.3ms,激光条纹中心提取精度均值为0.176pixel,标准差为0.119pixel,有效地保障了激光测量图像中激光条纹中心线的提取精度和鲁棒性。

An accurate and stable separation method for reflective areas in 3 D laser measurement images based on deep learning is proposed in this paper.Based on the deep learning model U-net network,the semantic segmentation of the reflective area and the laser line area in the 3 D laser measurement image is realized.The image semantic segmentation realizes the high-precision separation of different areas in the measurement image.The separated image of the reflective area can reduce the interference of the reflective on the centerline extraction.The research results show that the laser measurement image segmentation based on deep learning can more accurately achieve the extraction of laser stripes,while ensuring the stability of the extraction results.The segmentation time of a single laser measurement image is only 2.3 ms,the average extraction accuracy of the laser stripe center is 0.176 pixel,and the standard deviation is 0.119 pixel,which effectively guarantees the extraction accuracy and robustness of the laser stripe centerline in the laser measurement image.