Automatically compositing still images and landscape video sequences

We present an approach to deliver high-quality composites of panned, tilted, and zoomed landscape video sequences with computer-generated still images, targeted at large-scale construction projects. We propose a fully automatic camera-tracking algorithm to accurately composite computer-generated still images into panned, tilted, and zoomed video-sequence frames taken by hand or from a tripod. It provides an excellent means for a visual environmental assessment. First, we recover the camera parameters in fully automatic mode. Meanwhile, we initiate the matrix to match the coordinates between computer-generated objects and the video sequence´s first frame image by employing our algorithm and using the surrounding terrain map. We render computer-generated objects on a noninterlaced plane and erase the regions of their foreground by using image retouching software. Then, the algorithm automatically modifies the images and interlaces them to match each video-sequence frame by using the camera parameters. The images are also automatically composited with the video-sequence frames