ASPECT ANGLE ESTIMATION OF TARGETS IN FORWARD-LOOKING INFRARED IMAGESUSING THE MODEL-BASED VISION APPROACH

We develop a model-based vision algorithm to estimate the aspect angle of a target in a forward-looking infrared (FLIR) image. In the prepro cessing stage of the algorithm, a set of 3-D voxel-based models is cre ated using a CAD/CAM package. These models are rotated about the verti cal axis through a series of predetermined angles and then projected o nto the horizontal plane. This gives us a database library of 2-D imag es. We select as signature from a given FLIR image and attempt to matc h it with the various images in the given database library of images u sing the normalized cross-correlation method. The angle of rotation co rresponding to the image in the database library giving the best possi ble match is estimated to be the aspect angle of the signature (target ). We use an algebraic approach to represent images and the process in volves certain algebraic operations on the polynomials. An advantage o f the algebraic approach is that a high speedup in the run time is pos sible if the fast Fourier transform is used to compute the polynominal multiplications involved in the processing.