Wu Qingdong, Miao Jijun, Liu Zhaohui, Chang Jiaxiu
School of Civil Engineering, Qingdao University of Technology, Qingdao 266520, China.
Shandong Luqiao Group Co., Ltd., Jinan 250014, China.
Sensors (Basel). 2024 Sep 12;24(18):5926. doi: 10.3390/s24185926.
To improve the accuracy of camera calibration, a novel optimization method is proposed in this paper, which combines convex lens imaging with the bionic algorithm of Wolf Pack Predation (CLI-WPP). During the optimization process, the internal parameters and radial distortion parameters of the camera are regarded as the search targets of the bionic algorithm of Wolf Pack Predation, and the reprojection error of the calibration results is used as the fitness evaluation criterion of the bionic algorithm of Wolf Pack Predation. The goal of optimizing camera calibration parameters is achieved by iteratively searching for a solution that minimizes the fitness value. To overcome the drawback that the bionic algorithm of Wolf Pack Predation is prone to fall into local optimal, a reverse learning strategy based on convex lens imaging is introduced to transform the current optimal individual and generate a series of new individuals with potential better solutions that are different from the original individual, helping the algorithm out of the local optimum dilemma. The comparative experimental results show that the average reprojection errors of the simulated annealing algorithm, Zhang's calibration method, the sparrow search algorithm, the particle swarm optimization algorithm, bionic algorithm of Wolf Pack Predation, and the algorithm proposed in this paper (CLI-WPP) are 0.42986500, 0.28847656, 0.23543161, 0.219342495, 0.10637477, and 0.06615037, respectively. The results indicate that calibration accuracy, stability, and robustness are significantly improved with the optimization method based on the CLI-WPP, in comparison to the existing commonly used optimization algorithms.
为提高相机标定的精度，本文提出了一种新的优化方法，该方法将凸透镜成像与狼群捕食仿生算法（CLI-WPP）相结合。在优化过程中，将相机的内部参数和径向畸变参数作为狼群捕食仿生算法的搜索目标，将标定结果的重投影误差作为狼群捕食仿生算法的适应度评价标准。通过迭代搜索使适应度值最小的解来实现相机标定参数的优化目标。为克服狼群捕食仿生算法容易陷入局部最优的缺点，引入基于凸透镜成像的反向学习策略对当前最优个体进行变换，生成一系列与原个体不同的、具有潜在更好解的新个体，帮助算法摆脱局部最优困境。对比实验结果表明，模拟退火算法、张氏标定法、麻雀搜索算法、粒子群优化算法、狼群捕食仿生算法以及本文提出的算法（CLI-WPP）的平均重投影误差分别为0.42986500、0.28847656、0.23543161、0.219342495、0.10637477和0.06615037。结果表明，与现有的常用优化算法相比，基于CLI-WPP的优化方法显著提高了标定精度、稳定性和鲁棒性。
