An apparatus and method for preparing a composite image from a video data stream and for identifying changed features in two composite images. Frames from the video data stream are transformed to a projected 2D image, aligned with adjacent frames and overlapping areas are averaged to provide a higher apparent resolution. The composite image can be stored in real-time. As a second composite image is prepared of the same location at a later time, portions of the second image can be compared to corresponding portions of the stored image after the intensities of the images are equalized. Image areas whose absolute difference exceeds a threshold are again intensity equalized. Areas that are again above threshold can be flagged for further scrutiny, either by a human or by a machine that performs object recognition. In this way, composite video images of a scene can be prepared and compared in real-time.