基于机器视觉的轮椅床可行驶区域分割研究

Research on Caregiving Wheelchair-bed Free Space Segmentation Based on Machine Vision

针对现有具备护理功能的电动轮椅床在轮椅状态下的体积较大、驾驶较为困难等问题,提出一种基于机器视觉的轮椅床避障方案,采用视觉传感器采集环境信息,应用深度残差神经网络理论进行图像分割,检测出视场中的可行驶区域和障碍物。根据轮椅床实际运行环境特点,将公开数据集中的图片重新划分标签,采用重新划分标签后的数据集对分割模型进行训练。对训练后的模型进行可行驶区域分割试验,并与传统分割方法进行对比。结果表明,基于深度残差神经网络的分割方法解决了传统图像分割方法在地面和障碍物外观接近、2种不同地面的交界处以及光照不均匀等情况下易失效的问题,平均分割精度达到90%以上。

Aimed at the problems of big volume,hard driving of electric wheelchair-bed with nurse function under the station of wheelchair,a wheelchair-bed obstacle avoiding method based on machine vision was provided.The wheelchair-bed collected its surrounding environment information with visual sensors,deep residual neural network(DRNN)was used to segment the images to detect the free space and the obstacle.According to character of the space where the wheelchair-bed run,the public segmentation dataset was relabeled and then was used to train the deep residual neural network(DRNN).The model after training was tested in free space segregation,and the deep residual neural network(DRNN)was compared with traditional segmentation in experiments,the results showed that it solved lose efficacy of traditional image segmentation occurred in similar ground and obstacle,junction of different grounds,uneven illumination and so on.The average accuracy of the segmentation was over 90%.