Automated landmark detection is an important element of the examination and automatic analysis of 3D ultrasound images. The discovered 3D anatomical landmark points may simplify understanding of a 3D heart orientation and track anatomical structures during the heart cycle. This has further impact on the accuracy of diagnostics and reliability of the final diagnosis. The algorithm presented in this paper applies landmark detection in three perpendicular planes of the 4D VOLDICOM dataset with the fourth dimension representing time. The detection is performed using a cascade of two trained AdaBoost classifiers relying on an extended set of Haar-like features and exploits a manually annotated dataset of 2D ultrasound cardiac images in each plane. The resulting method is able to effectively detect the left ventricle apex and the mitral valve center in the two-chamber heart views with an accuracy comparable to those of the manual detection and other automatic methods presented in the literature.