ESTIMATING CAMERA MOTION THROUGH VISUAL TRACKING IN LOW CONTRAST HIGH MOTION SINGLE CAMERA SYSTEMS

Systems and methods are provided that track camera motion from image and sensor data in single-camera, low contrast and high-motion systems. Camera motion is estimated through dense visual tracking using image and sensor data. Motion sensor data from a wearable motion sensor worn by a human is used to determine initial camera motion parameters. Image data from a thermal imaging camera outputting low contrast video frames is used for motion tracking. The camera motion in the frame is represented by a translation and a rotation of the camera through an environment. The frames are down-sampled to generate image pyramid of frames of progressively lower resolution. A hierarchical homography optimizing approach is described. A homography is optimized across each resolution level beginning with the lowest resolution frames. A modified translation and rotation displacement of the camera is determined based on the optimized homography.