We aim at investigating arterial diseases at early stage, by assessing the longitudinal (i.e. in the same direction as the blood flow) motion of the intima-media complex. This recently evidenced phenomenon has been shown to provide relevant and complementary information about vascular health. Our method assesses the longitudinal and radial motion from clinical in vivo B-mode ultrasound sequences. To estimate the trajectory of a selected point during the cardiac cycle, we introduce a block matching method that involves a temporal update of the reference block using a pixel-wise Kalman filter. The filter uses the initial gray-level of the pixel as control signal to avoid divergence due to cumulating errors. The block and search-window sizes are adapted to the tissue of interest. The method was evaluated on image sequences of the common carotid artery, acquired in 57 healthy volunteers and in 25 patients at high cardiovascular risk. Reference trajectories were generated for each sequence by averaging the tracings performed by three observers. Six different computerized techniques were also compared to our method. With a pixel size of 30 μm, the average absolute motion estimation errors were 84 ± 107 μm and 20 ± 19 μm for the longitudinal and radial directions, respectively. This accuracy was of the same order of magnitude as the inter- and intra-observers variability, and smaller than for the other methods. The estimated longitudinal motion amplitude was significantly reduced in at-risk patients compared with healthy volunteers (408 ± 281 μm vs. 643 ± 274 μm, p < 0.0001). Our method can constitute a reliable and time-saving technique to investigate the arterial stiffness in clinical studies, in the objective to detect early-stage atherosclerosis.