基于视觉里程计的单目红外视频三维重建

3D reconstruction of monocular infrared video based on visual odometer

针对红外视频的特点,提出了一种基于直接法和稀疏法视觉里程计的单目红外视频三维重建方法。该方法首先通过对红外热像仪标定获得热像仪内参,然后构建直接法和稀疏法视觉里程计模型,视觉里程计前端执行帧管理和点管理的任务,利用滑动窗口并借助高斯-牛顿迭代对总光度误差进行优化,计算出直接法和稀疏法视觉里程计模型所依赖的所有变量,完成定位热像仪和建图的任务。通过实验证明了该方法能够实时实现对单目红外视频进行三维重建。

Aiming at the characteristics of infrared video,a three-dimensional reconstruction method of monocular infrared video based on direct and sparse visual odometer is proposed. In this method,the internal parameters of the thermal imager are obtained by the calibration of the infrared camera,and then the direct and sparse visual odometer model is constructed. Visual odometer front-end performs frame management and point management tasks. The total photometric error is optimized by the sliding window and the Gauss-Newton iteration,and all the variables that the direct and sparse visual odometer model needs are calculated. Finally,the tasks of locating the thermal imager and plotting are fullfilled. Experimental results show that the proposed method can achieve the three-dimensional reconstruction of monocular infrared video in real time.