This paper describes several innovative algorithms for the on-line estimation of the image Jacobian based on both the incorporation of the epipolar geometry properties, i. e. the fundamental matrix, and the definition of a reliability factor. Reliability factor depends on both displacements in image feature space and movements in joints space. These new algorithms have been applied successfully in positioning and tracking tasks of a uncalibrated joint system of three d.o.f. with two fixed weakly calibrated cameras. Several tests performed show that the algorithm based on the fundamental matrix is even more robust in the presence of noise than those base on recursive methods.