Implementing automatic visual tracking for underwater robots often remains a great challenge. In this paper, we demonstrate an active visual tracking case with a free-swimming robotic fish by use of a combination of TI DM3730-centered embedded hardware and software. Specifically, automatic recognition algorithms mainly involving color-based identification and shape-based identification are proposed. In parallel with the enhanced target recognition, an improved real-time location method based on the weighed histogram representation of the target model that synthesizes the CAMSHIFT algorithm and the Kalman filter, is put forward. After that, a fuzzy logic controller based on the CPG model is employed to generate control signals related to the desired target. Aquatic experiments validate the effectiveness of the proposed algorithms.