Testing a dynamic algorithm for image matching in a CCD sequence acquired by a low cost MMS

This paper deals with the testing of a matching algorithm to search homologous points onto images consecutively acquired by the CCD of our “simplified low cost MMS”. In such a prototype, a GPS receiver, an odometer, only one CCD camera, and a pc are employed and fixed onto a vehicle. The expensive INS sensor, only yielding relative positioning/attitude if used alone, is therefore not used; to execute the relative orientation, consecutive images acquired by a merely monoscopic-imaging sensor are exploited. The paper goal is to test the numerical performances/robustness of the implemented algorithm, automatically detecting the homologous point in the “second” image, once a point is (manually) chosen in the “first” image. The analytical innovation is the employment of a Kalman filter dynamic model to improve the convergence of the iterative solution. The constraint given by the “epipolar line” is engaged, but it is used as a “geometrical state equation” for the filter. In this way, the epipolar constraint is fulfilled in stochastic sense, allowing the weight setting of such a condition. It will strongly restrict the unknown solution if the second image is already oriented, while it will be a weaker constraint when the orientation is only approximately known. The implementation of the matching model in a software tool has been accomplished by using MATLAB program, and well promising results on real imaging data have been obtained. Within 12-14 iterations, stables values for the geometric/radiometric matching unknown are achieved. At the same time, the resulting shape of the patch window on the second image, centred on the searched homologous point, quickly becomes stationary no more moving from the actual (unknown) solution.