«上一篇
 文章快速检索 高级检索

 智能系统学报  2021, Vol. 16 Issue (6): 1045-1055  DOI: 10.11992/tis.202103014 0

### 引用本文

ZHU Qidan, LI Xiaotong, ZHENG Tianhao. Research on real-time vision measurement algorithm of shipborne aircraft pose[J]. CAAI Transactions on Intelligent Systems, 2021, 16(6): 1045-1055. DOI: 10.11992/tis.202103014.

### 文章历史

Research on real-time vision measurement algorithm of shipborne aircraft pose
ZHU Qidan , LI Xiaotong , ZHENG Tianhao
College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin 150001, China
Abstract: Real-time detection of carrier aircraft pose is of great significance to the movement control, trajectory planning, and collision avoidance of carrier-based aircraft on deck. The traditional scheduling of carrier aircraft mainly relies on the manual judgment of the position and heading angle of the carrier-based aircraft. Traditional methods fail to obtain accurate data and readily cause collisions because of operator negligence and fatigue. Therefore, a real-time visual measurement algorithm for the pose of carrier aircraft is proposed. First, the carrier aircraft is identified and segmented based on the you only look once version 4 (YOLO-V4) network and the Canny edge extraction algorithm. Then, an innovative algorithm for wireframe template matching is proposed, and the best pose is obtained by calculating the matching degree between the contour of the carrier aircraft and the wireframe template. The algorithm meets the real-time requirements through parallelization and graphics processing unit acceleration, and the test is completed in the 1:70 and 1:14 physical simulation environments. The results show that the recognition rate of this algorithm is >95%, the accuracy is within 0.7° and 8 mm, and the speed can reach 8 Hz.
Key words: machine vision    shipborne aircraft pose    deep learning    target detection    edge extraction    wireframe matching    GPU acceleration    target segmentation

1 舰载机位姿实时测量系统

2 基于深度学习的目标检测与分割

3 线框模板匹配位姿测量算法 3.1 线框模板匹配算法

3.2 Canny边缘提取算法及其优化

1)高斯滤波

 ${g_\sigma }(m,n) = \frac{1}{{\sqrt {2{{\text{π}}}{\sigma ^2}} }}{{\text{e}}^{ - \frac{{{m^2} + {n^2}}}{{2{\sigma ^2}}}}} \cdot f(m,n)$ (1)

2)计算梯度幅值与方向

 $G(m,n) = \sqrt {{g_x}{{(m,n)}^2} + {g_y}{{(m,n)}^2}}$ (2)
 $\theta = \arctan \frac{{{g_y}(m,n)}}{{{g_x}(m,n)}}$ (3)

3)过滤非最大值

4)上下阈值过滤

3.3 舰载机线框模板的建立

3.4 基于鸟瞰算法的多目标去重与初始解估计

 ${Z_c}\left[ {\begin{array}{*{20}{c}} u \\ v \\ 1 \end{array}} \right] = \left[ {\begin{array}{*{20}{c}} {{f_x}}&0&{{u_0}} \\ 0&{{f_y}}&{{v_0}} \\ 0&0&1 \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {{R}}&{{T}} \end{array}} \right]\left[ {\begin{array}{*{20}{c}} {{X_w}} \\ {{Y_w}} \\ {{Z_w}} \\ 1 \end{array}} \right] = {{\boldsymbol{M}}_{{1}}}{{\boldsymbol{M}}_{{2}}}X$ (4)

3.5 匹配度算法的选择与优化

1)点投影在线段上；

2)点投影不在线段上。

3.6 算法并行化及GPU加速

4 算法验证与性能测试 4.1 硬件环境搭建与算法测试软件编写 4.1.1 硬件环境搭建

4.1.2 算法测试软件编写

1)相机视野实时检测；

2)相机内外参标定；

3)实时Canny边缘提取结果显示；

4)舰载机位姿数据显示；

5)舰载机位姿图像化显示。

4.2 算法性能测试

4.2.1 鲁棒性测试

1)光照因素

2)多舰载机目标分割范围重叠

4.2.2 实用性测试

4.2.1节所进行的算法测试是在室内进行的，无外界强烈阳光的影响，并且对应的舰载机模型都是1∶72的模型。而在实际的应用中，一方面，应用场景是在海上，本文提出的舰载机位姿解算算法必然会受到室外光照的影响；另一方面，实际应用场景中对应的待测目标是用于作战的舰载机，其大小并不是1∶72的舰载机模型的大小。因此，本文提出的算法要想具有实用性，能应用到实际的场景中，一方面必须得适应室外的阳光，另一方面还必须具有能适应不同大小舰载机模型的能力。为了验证本文提出的算法具有实用性，我们将实验环境换到具有强烈阳光的室外，并将舰载机模型换为1∶14的歼-12模型，得到测试结果如图16所示。

4.2.3 精度测试

4.2.4 实时性测试