Stereo Visual Odometry (with Bundle Adjustment)
github: https://github.com/Juhyung-L/stereo_visual_odometry Visual odometry is the process of estimating a camera's position by processing the individual images taken from a moving camera. It can be performed with different types of cameras: monocular (single camera), stereo (two cameras), depth (camera with depth info). In this blog, I will explain the inner workings of a visual odometry system for a stereo camera setup. Figure 1: My implementation of stereo visual odometry. A stereo camera is a pair of cameras facing the same direction that are separated by a known distance. Figure 2: Stereo camera setup. Unlike a monocular camera, the stereo camera can estimate the depth of a pixel on an image using a technique called triangulation. This is similar to how humans can sense how far away an object is using both eyes. The system also includes bundle adjustment, a non-linear error minimization technique used to reduce the accumulation of error in the pose estimates. I hope to see i