Contactless respiration monitoring using Doppler radar is an important technology for healthcare applications. The radar measures small displacements of the body surface. In this study, we propose a new algorithm to separate multiple targets placed closely together at the same range but at different lateral positions using ultra-wideband array radar and the Capon method. The Capon method, which is an adaptive beamforming - technique, assumes that arrival echoes are not correlated. However, echoes from the human body should be correlated and this reduces system performance. We improve the performance - by introducing two different diagonal loading factor values for direction of arrival estimation and weight vector calculation. In an experimental study, the proposed method separates the echoes from two patients spaced at a lateral distance of approximately 70 mm. The estimated displacement error when using the proposed method is less than 0.13 mm, while that of the conventional method is 2.7 mm.