Road Survey by Kalman Filter Rectification of Image Sequences Acquired with a Monoscopic Low-cost Mms

This paper proposes a method for a mosaic digital representation of a road by means of a Kalman filter based rectification of a monoscopic image sequence acquired with a low-cost MMS without INS sensors. The frontal images of the road acquired by a pointing ahead-down camera are rectified onto the road surface by means of a homographic transformation among image and ground planes. As basic idea, the “external/homographic” parameters constitute the unknown “state vector” of a Kalman filter pseudodynamic model, describing its evolution with respect to the covered distance, with observation equations given from image coplanarity and collinearity conditions and GPS kinematic measures. To make easy the relative orientation among images, an original Kalman filter matching is exploited: once selected some lane tract points in the “first” image, the homologous points in the “second” image are so automatically detected. Exploiting only near points in the lower part of the images, a high-scale optical model with a strong geometrical consistency is achieved: therefore, a reliable rectification of the image sequence can be expected. This method has been implemented in a Matlab language program making possible the rectification of the image sequence by a digital resampling with automatic mosaicking onto finite planes modeling the road surface. First numerical experiments with real data of our MMS prototype have given satisfactory results here described and analyzed.