Ultrasound imaging is a helpful aid for diagnosis in many medical specialities. In order to enhance the information available to the examiner, an upcoming approach is to compound the sequence of video images into a single 3D volume, which needs a real-time tracking of the transducer. However, errors in the positioning occur, and the compounded image becomes degraded. We propose a method which automatically corrects the positioning error, by means of a registration step between the live ecography and the compounded volume. The algorithm has been tested on sequences taken on an in-vitro human brain.