An active vision system with two cameras has to be calibrated on-line to perform stereo vision applications. A calibration of a binocular system is possible using a small number of homologous features in both images. The correspondence problem can be resolved by using traditional template matching algorithms but it has a high computational cost. In this paper, some dynamic programming aspects are introduced to deal with template matching in order to reduce the execution time. Some experimental results demonstrate their viability and suggest other lines to improve its performance.