We are developing a new system called Computer-Aided Tracking and Motion Analysis with Ultrasound System (CAT & MAUS) to dynamically describe joint kinematics for pathology research on musculoskeletal conditions. It is essential to have computer-aided bony structure tracking in ultrasound (US) sequences in such a system for practical use. However, unlike CT or MRI imaging, the appearance of bony landmarks in US imagery can be distorted by the image formation process. In this paper, a target patch tracking approach is presented for tracking ultrasound landmarks. The tracker is based on compressed intensity features which are stored in k-dimensional (K-D) trees for fast and discriminatory image patch searches. Dynamic programming is used to find the optimal track through the ultrasound video. We find that our method is more accurate than the mean-shift and the KLT trackers for bony structures, especially with large movements.