Automatic vehicle image extraction based on spatio-temporal image analysis

Presents a robust method for extracting images of laterally moving vehicles from ITV image sequences. The information of occlusion, velocity and length of the vehicle is first extracted from the spatio-temporal image, and the background image is updated periodically. Overlapping images of other vehicles are then removed from each scene and a subtraction image sequence is superimposed. The new method has the following characteristics: (1) one can separate and extract images of two crossing vehicles from a sequence of images. (2) the complete image of a large or complex vehicle can be successfully extracted under typical daylight variations. In order to test the applicability of the extracted images, the authors classify passenger car types. Experiments with real data confirm the effectiveness of the method