矿井巷道复杂场景灭点检测方法

程健; 王瑞彬; 郁华森; 闫鹏鹏; 王凯

doi:10.13272/j.issn.1671-251x.2021040097

矿井巷道复杂场景灭点检测方法

doi: 10.13272/j.issn.1671-251x.2021040097

程健^{1, 2},
王瑞彬³,
郁华森⁴,
闫鹏鹏^{1, 2, 3},
王凯^{1, 2}

1.
煤炭科学研究总院矿山大数据研究院，北京100013
2.
煤炭资源高效开采与洁净利用国家重点实验室，北京100013
3.
中国矿业大学信息与控制工程学院，江苏徐州221116
4.
首都医科大学生物医学工程学院，北京100069

基金项目:

辽宁省自然科学基金资助项目(2020-KF-22-02)

北京市科技计划项目(Z201100004520015)

中国煤炭科工集团有限公司科技创新创业资金专项重点项目(2019-2-ZD002)

详细信息

中图分类号: TD67
计量
- 文章访问数: 108
- HTML全文浏览量: 18
- PDF下载量: 11
- 被引次数: 0
出版历程
- 刊出日期: 2021-06-20

Vanishing point detection method in complex environment of mine roadway

CHENG Jian^{1, 2},
WANG Ruibin³,
YU Huasen⁴,
YAN Pengpeng^{1, 2, 3},
WANG Kai^{1, 2}

1.
Research Institute of Mine Big Data, China Coal Research Institute, Beijing 100013, China
2.
State Key Laboratory of Coal Mining and Clean Utilization, Beijing 100013, China
3.
School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, China
4.
School of Biomedical Engineering, Capital Medical University, Beijing 100069, China

摘要

摘要: 通过检测识别图像中的灭点位置可用来辅助矿井巷道移动机器人进行自主导航。针对现有的灭点检测方法在光照条件差、结构化信息不足的矿井巷道场景下误差较大的问题，提出了一种矿井巷道复杂场景灭点检测方法。首先，对图像进行缩小、滤波、灰度化等预处理，以大幅减少计算量，较好地保留直线特征；然后，采用直线检测算法对图像进行直线检测，引入直线长度阈值和平均梯度约束分别剔除长度小的干扰直线和图像中由阴影产生的干扰直线，并采用块匹配算法生成图像的块运动轨迹直线；最后，将剔除干扰后的直线和块运动轨迹直线转换为参数空间中的样本点，采用局部异常因子算法求出每个样本点的离群因子值，并将样本点的离群因子值和对应直线长度共同作为衡量样本点重要性的标准，据此设计加权回归算法的权函数，从而得到灭点的最佳估计。在矿井巷道数据集与公共数据集上的实验结果表明，与基于边缘的灭点检测方法和基于深度学习的灭点检测方法相比，本文方法对光照变化有较强的鲁棒性，在光照条件差、缺乏直线信息的复杂场景中具有更高的精度，且实时性优于基于深度学习的灭点检测方法，能够更好地满足矿井巷道机器人导航需求。
- 矿井巷道 /
- 灭点检测 /
- 机器人自主导航 /
- 直线检测 /
- 平均梯度约束 /
- 块匹配
Abstract: By detecting and identifying the vanishing point position in the image, it is able to assist mobile robots in mines roadways for autonomous navigation. The existing vanishing point detection methods have large errors in the mine roadway with poor lighting conditions and insufficient structured information. In order to solve the above problems, a vanishing point detection method in complex environment of mine roadways is proposed. Firstly, the image is pre-processed by reducing, filtering, graying, etc. This method can reduce the calculation amount significantly and the straight line characteristics can be better preserved. Then, the straight line detection algorithm is used to detect the straight line of the image. The straight line length threshold and the average gradient constraint are introduced to eliminate the interference line with small length and the interference line generated by shadows in the image respectively. Moreover, the block matching algorithm is used to generate the block motion trajectory straight line of the image. Finally, the straight lines after removing the interference and the block motion trajectory straight lines are converted into sample points in the parameter space. The outlier factor value of each sample point is calculated by the local anomaly factor algorithm, and the outlier factor value of the sample point and the length of the corresponding straight line are used as the criteria to measure the importance of the sample points. On this basis, the weight function of the weighted regression algorithm is designed to obtain the best estimate of the vanishing point. The experimental results on the mine roadway data set and public data set show that compared with the edge-based vanishing point detection method and the deep learning-based vanishing point detection method, the method in this paper has stronger robustness to light changes. It has higher accuracy in complex environment with poor lighting conditions and lack of straight line information, and has better real-time performance than the vanishing point detection method based on deep learning. This method can better meet the needs of mine roadway robot navigation.
- mine roadway /
- vanishing point detection /
- autonomous robot navigation /
- straight line detection /
- average gradient constraint /
- block matching