«上一篇
 文章快速检索 高级检索

 哈尔滨工程大学学报  2019, Vol. 40 Issue (4): 844-850  DOI: 10.11990/jheu.201709095 0

### 引用本文

NIU Jie, BU Xiongzhu, QIAN Kun. A method of extracting natural landmarks for mobile robot navigation[J]. Journal of Harbin Engineering University, 2019, 40(4), 844-850. DOI: 10.11990/jheu.201709095.

### 文章历史

1. 常州信息职业技术学院 电子与电气工程学院, 江苏 常州 213164;
2. 南京理工大学 机械工程学院, 江苏 南京 210094;
3. 东南大学 自动化学院, 江苏 南京 210096

A method of extracting natural landmarks for mobile robot navigation
NIU Jie 1,2, BU Xiongzhu 2, QIAN Kun 3
1. School of Electrical and Electronic Engineering, Changzhou College of Information Technology, Changzhou 213164, China;
2. School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China;
3. School of Automation, Southeast University, Nanjing 210096, China
Abstract: Owing to the shortage of artificial landmarks in robot localization and navigation applications, a method of extracting a significant landmark is presented on the basis of frequency domain characteristics.First, this method used the image entropy technique to adaptively select the factor to smooth the image.Then, salience maps of the three-channel color space were obtained by the frequency domain saliency method in the opposite color space.Thus, weighted fusion was conducted.The landmark must be consistent and noise should be reduced.Optimized K-means image clustering method was then used to obtain masked final landmarks.The natural landmarks available for robot navigation applications were selected.The experiments show that the pixels extracted by the visual feature reach an average detection rate of 80%.Furthermore, the proposed method has high reproducibility indoors compared with direct matching of characteristic operators.Finally, practical robot navigation based on the natural landmark validates the effectiveness of the method.
Keywords: visual attention    image clustering    image segmentation    mobile robot    landmark detection

1 基于频域特性的显著区域提取

 Download: 图 1 显著路标提取算法框架 Fig. 1 Illustration of the main phases of algorithm
1.1 图像自适应平滑预处理

 $\mathop {{\rm{min}}}\limits_f \sum\limits_p {{{\left( {{f_p} - {g_p}} \right)}^2}} + {\rm{ \mathsf{ λ} }} \cdot c\left( f \right)$ (1)

 ${H_{{R_i}}} = - \sum\nolimits_{v = 0}^{255} {{p_v}{\rm{ln}}{p_v}}$ (2)

1.2 候选显著区域提取

1) 区域所占图像的比例尺寸上有一定的限制，不能太小，也不能过大;

2) 能够均匀突出显示整个路标区域;

3) 具有明确的对象边界;

4) 忽略由纹理、噪声引起的高频块效应;

5) 有效输出全分辨率显著图。

 $\begin{array}{l} {D_{{\rm{oG}}}}\left( {x, y} \right) = \frac{1}{{2{\rm{ \mathsf{ π} }}}}\left[ {\frac{1}{{\sigma _1^2}}{{\rm{e}}^{ - \frac{{\left( {{x^2} + {y^2}} \right)}}{{2\sigma _1^2}}}} - \frac{1}{{\sigma _2^2}}{{\rm{e}}^{ - \frac{{\left( {{x^2} + {y^2}} \right)}}{{2\sigma _2^2}}}}} \right] = \\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;G(x, y, {\sigma _1}) - G(x, y, {\sigma _2}) \end{array}$ (3)

 $\begin{array}{l} \sum\limits_0^{N - 1} {G\left( {x, y, {\rho ^{n + 1}}\sigma } \right)} - G(x, y, {\rho ^n}\sigma ) = \\ \;\;\;\;\;\;\;\;G(x, y, \sigma {\rho ^N}) - G\left( {x, y, \sigma } \right) \end{array}$ (4)

 $S\left( {x, y} \right) = \left\| {{\mathit{\boldsymbol{I}}_u} - {\mathit{\boldsymbol{I}}_{{\rm{wht}}}}\left( {x, y} \right)} \right\|$ (5)

 $\left\{ \begin{array}{l} I = \left( {R + B + G} \right)/3\\ {R_G} = R - G\\ {B_Y} = B - \left( {R + G} \right)/2 \end{array} \right.$ (6)

 Download: 图 2 对立颜色空间各通道显著图 Fig. 2 Saliency computation in the opponent color space

1) 分别计算出IRGBY通道的显著图。

2) 分别计算出各个通道显著图像的平均显著值，并统计超过该平均值的像素数量百分比，分别用NINRGNBY表示。

3) 计算各通道的加权系数，计算公式为：

 $w = \left\{ \begin{array}{l} {\rm{min}}\left( {1, 1 - {{\left( {N - r} \right)}^2}} \right), \;\;\;\;\;\;\;\;\;\;\;\;N \le r\\ {\rm{max}}\left( {0, 0.5 - {{\left( {N - r} \right)}^{1/2}}} \right), \;\;\;\;\;\;其他 \end{array} \right.$ (7)

4) 计算最终的图像显著图，计算公式为：

 ${S_{gs}} = {\rm{Norm}}\left( {\sum {{W_I}{S_I}} , {W_{RG}}{S_{RG}}, {W_{BY}}{S_{BY}}} \right)$ (8)

2 基于聚类算法的显著区域分割

 ${V_k} = \frac{1}{{\left| {{r_k}} \right|}} = \sum\limits_{i, j \in {r_k}} {{m_{i, j}}}$ (9)

 $T = \frac{1}{{W \times H}}\sum\limits_{x = 0}^{W - 1} {\sum\limits_{y = 0}^{H - 1} {S\left( {x, y} \right)} }$ (10)

3 实验结果与分析

 Download: 图 4 活动室场景及其显著区域示例 Fig. 4 Example of the activity room and its salient areas
 Download: 图 5 机器人运动中显著区域像素检测率 Fig. 5 Salient pixels detection rate during the robot motion