Interdisciplinary Graduate School of Science and Engineering, Tokyo Institute of Technology, Yokohama 226-8503, Kanagawa, Japan.
School of Science and Engineering, Tokyo Institute of Technology, Yokohama 226-8503, Kanagawa, Japan.
Sensors (Basel). 2021 Oct 24;21(21):7048. doi: 10.3390/s21217048.
Direct visual odometry algorithms assume that every frame from the camera has the same photometric characteristics. However, the cameras with auto exposure are widely used outdoors as the environment often changes. The vignetting also affects the pixel's brightness on different frames, even if the exposure time is fixed. We propose an online vignetting correction and exposure time estimation method for stereo direct visual odometry algorithms. Our method works on a camera that has a gamma-like response function. The inverse vignetting function and exposure time ratio between neighboring frames are estimated. Stereo matching is used to select correspondences between the left image and right image in the same frame at the initialization step. Feature points are used to pick the correspondences between different frames. Our method provides static correction results during the experiments on datasets and a stereo camera.
直接视觉里程计算法假设相机的每一帧都具有相同的光度特性。然而,自动曝光相机在户外被广泛使用,因为环境经常变化。即使曝光时间固定,渐晕也会影响不同帧上的像素亮度。我们为立体直接视觉里程计算法提出了一种在线渐晕校正和曝光时间估计方法。我们的方法适用于具有类伽马响应函数的相机。估计了相邻帧之间的逆渐晕函数和曝光时间比。在初始化步骤中,立体匹配用于在同一帧的左图像和右图像之间选择对应关系。特征点用于在不同帧之间选择对应关系。我们的方法在数据集和立体相机上的实验中提供了静态校正结果。