This paper presents a system that creates and navigates an unlimited-size mosaic with geographical information. The input is a sequence of airborne images with or without telemetry data, and the output is a mosaic with a combined geographical coordinate layer inherited from the input images. Rather than registering input images with an orthoimage, which is popular in existing applications, the proposed system only takes use of telemetry data as prior information. The airborne images embedded with geo-information are pair-wise registered, based on image feature correspondence. We extract feature points and form a modified EDGE-based descriptor for image registration. Subsequently, the geographical coordinate layers derived from the telemetry data stream are fused using a registration matrix computed from the previous step. However, due to the unreliability of the telemetry data, the new geodetic coordinate layer might be inconsistent with the image coordinate layer and therefore requires rectification to minimize the squared error between the mosaic coordinate layer and the warped geographical coordinate layer. The above process is incorporated into a cluster framework so that the output mosaic is extensible to an infinite size. That is, once the current mosaic size has expanded beyond computer memory limitations, the image is saved to a database. Its spatial relationship with respect to the world coordinate system is also saved to the database so that the system can navigate the collection of image mosaic data by querying the spatial database and retrieving the relevant mosaics. This method is especially suitable for video sequences spanning large regions, such as surveillance video from a micro UAV. Results with real-world UAV video are provided to demonstrate the performance of the proposed system.
This paper presents a system for creating a mosaic image from a sequence of images with moving objects present in the scene. This system first uses SIFT-based image registration on the entire image to obtain the initial global projection matrix. After image segmentation, the global motion model is applied to each region for evaluation. The transformation matrix is refined for best projection in each region, and a more precise global transformation matrix is calculated based upon local projections on majority of coherent regions. As a consequence this method is robust to disturbances to the projection model induced by moving objects and motion parallax. In the image blending stage, pixels in coherent regions are weighted by their distances from the overlapping edges to achieve a seamless panorama, while heterogeneous regions are cut and pasted to avoid ghosting or blurring. The most recent information regarding location, shape, and size of the moving foreground objects is therefore reflected in the panorama. Constructed mosaics are presented to demonstrate the performance and robustness of the proposed algorithm.
This paper presents a method for localizing noise-corrupted areas in quality degraded video frames, and for
reducing the additive noise by utilizing the temporal redundancy in the video sequence. In the proposed algorithm,
the local variance of each pixel is computed to obtain the spatial distribution of noise. After adaptive
thresholding, region clustering, and merging, the corrupted areas of highest energy are detected. Due to the high
temporal redundancy in the video sequence, the corrupted information can be compensated by overlapping the
corrupted regions with the appropriate regions from adjacent video frames. The corresponding pixel locations
in the adjacent frames are computed by using image registration and warping techniques. New pixel values
are calculated based upon multi-frame stacking. Pixel values in the adjacent frames are weighted according to
registration errors, whereas the values in the noisy frame are evaluated according to local variance. Knowing
the location of the local noise enables the denoising process to be much more specific and accurate. Moreover,
since only a portion of the frame is processed, as compared to standard denoising methods that operate on the
entire frame, the details and features in other areas of the frame are preserved. The proposed scheme is applied
to UAV video sequences, where the outstanding noise localization and reduction properties are demonstrated.
This paper describes an automatic video target tracking system that operates on the panoramic image provided by an image mosaicking preprocessing stage. In the mosaic preprocessing stage, a feature-based algorithm is applied to obtain the underlying homography between consecutive frames in a video sequence. With the first frame in the sequence chosen as the base image plane, subsequent frames are warped and merged into a panoramic scene for the video tracking stage. The tracking algorithm calculates the motion vector for each block in a warped frame by comparing it with the panoramic image, and those exceeding the dominant background motion can be considered as blocks belonging to potential moving foreground objects. Image segmentation is then used to recover the boundaries of the foreground objects. After fusing the labeled boundaries with the motion vector information, the potential targets, as well as their feature vectors, are identified. The feature vectors include information pertaining to location, size, and optical characteristics, and are input into a sub-tracker for record keeping. The input to the proposed system is a video stream from a single camera.