基于改进图像增强的低照度场景视觉惯性定位方法

Visual-inertial localization method in low-light scene based on improved image enhancement

为提升视觉惯性导航系统在低照度场景下的定位精度,提出一种结合图像增强技术的视觉惯性定位方法。根据不同曝光图像的直方图确定相机响应模型,通过曲线拟合确定模型参数。利用非线性优化得到低照度图像的照明图以及曝光率矩阵,根据相机响应模型对低照度图像进行预处理。使用光流法进行特征追踪,将视觉误差、IMU误差以及先验误差作为约束,构建紧耦合优化模型,从而实现更精确的位姿估计。最后使用车载设备采集的真实数据对所提方法进行了评估,实验结果表明:所提方法能有效提升视觉惯性导航系统在低照度场景下的定位精度,相比于无图像增强的方法,定位精度提高了25.59%;相比于改进前的图像增强方法,定位精度提高了6.38%。

In order to improve the localization accuracy of the visual-inertial navigation system in the low-light scene,a visual-inertial localization algorithm combined with image enhancement technology is proposed.The camera response model is determined according to the histograms of different exposure images,and the model parameters are determined by curve fitting.The illumination map and exposure matrix of low-light images are determined by nonlinear optimization,and the low-light images are preprocessed according to the camera response model.The optical flow method is used for feature tracking,and the visual error,inertial measurement unit(IMU)error and prior error are used as constraints to construct a tightly-coupled optimization model,so as to achieve more accurate pose estimation.Finally,the method is evaluated using real data collected by on-board equipment.The experimental results show that the proposed method can effectively improve the localization accuracy of the visual-inertial navigation system in the low-light scene.Compared with the method without image enhancement,the localization accuracy increased by 25.59%.Compared with the method before improvement,the localization accuracy increased by 6.38%.