【作者】 汤伟； 刘思洋； 陈景霞； 高涵； 陶倩；

【Author】 TANG Wei;LIU Si-yang;CHEN Jing-xia;GAO Han;TAO Qian;College of Electrical and Information Engineering,Shaanxi University of Science & Technology;College of Mechanical and Electrical Engineering,Shaanxi University of Science & Technology;

【机构】 陕西科技大学电气与信息工程学院； 陕西科技大学机电工程学院；

【摘要】 针对中小型水域漂浮垃圾清理困难问题,提出基于计算机视觉的水面目标分割与识别方法,利用计算机技术为水面垃圾清理机器人提供指导。首先,使用Mean-shift法综合像素点的值域特征与空间域信息,对图像进行聚类平滑,同时结合OTSU法解决复杂背景下水面目标的分割;然后,设计用于描述水面目标的颜色、几何、纹理特征参数,根据5种水面目标在三类特征上差异的显著性构建分步识别流程;最后,建立用于多分类的支持向量机(SVM)模型,按识别流程进行识别。实验结果表明,相比于传统OTSU法和Niblack法,本文方法具有更好的分割效果,5种水面目标均可从复杂背景中提取出来,并可被快速识别,平均识别准确率为93%,平均漏报率和误报均未超过7%,与用所有特征参数进行直接识别相比,本文提出的分步识别方法在准确率、可靠性和执行时间上均有一定优势,可为水面机器人的工作提供技术支撑。更多还原

【Abstract】 Aiming at the difficulty of garbage cleaning in small or medium-sized waters, a water surface target segmentation and recognition method based on computer vision was proposed. Computer technology was used to provide guidance for surface cleaning robot. Firstly, the Mean-shift method is used to synthesize the value information and spatial domain information of the pixel, and the image is clustered and smoothed. OTSU method is combined to solve the segmentation of the water surface target under complex background. Then, the color, geometry and texture parameters are designed to describe the water surface target. And a step-by-step recognition process is constructed based on the saliency of the five types of surface targets on the three types of features. Finally, the support vector machine(SVM) model for multi-classification is established. and targets is identified according to the process. The experimental results show that compared with the traditional OTSU and Niblack method, the proposed method has better segmentation effect. The five water surface targets can be extracted from the complex background and can be quickly identified. The average recognition accuracy is 93%. The average missing report rate and false report rate are 7% lower. Compared with the direct identification of all characteristic parameters, the step-by-step identification method has certain advantages in accuracy, reliability and execution time. It can provide technical support for the work of surface robots.更多还原