Multi-target tracking with low-cost radars on USVs is challenging due to interference, vessel vibrations, and occlusions, making single-sensor methods inadequate. This study considers multiple USVs collaboratively tracking multiple targets using only latitude and longitude from low-cost radars. First, we associate radar detections to targets via the Hungarian algorithm. Second, because noise and clutter prevent clear cluster separation, we introduce Dilate K-means: over-cluster the data, then dilate clusters to identify the optimal count. Finally, we employ an adaptive Kalman filter that adjusts measurement and process noise covariances online. Experiments confirm that our approach delivers robust, accurate multi-target tracking in complex marine environments.