Mobile SAM-based Motion Target Detection

The study introduces a motion target detection algorithm that aims to effectively detect unconventional targets such as irregularly shaped unmanned aerial vehicles (UAVs), unmanned ground vehicles (UGVs), and specialized robots. Initially, Mobile SAM (Mobile Segment Anything Model) is utilized for object segmentation, extracting mask information for all entities within a frame. Subsequently, the centroid coordinates extracted from the mask data are used as input for the K-Nearest Neighbors (KNN) algorithm. By integrating trajectory prediction and data fusion technologies, the objective of motion target detection is achieved. Experiments make use of publicly available datasets for UAV detection and tracking in order to design and validate the algorithm. The results demonstrate a 6.73% improvement in detection accuracy, an 8.63% improvement in recall rate, and a 7.66% improvement in F1 score compared to the modified three-frame differential motion target detection algorithm. This research makes significant contributions by addressing the limitations of traditional motion target detection, reducing the algorithm’s dependence on datasets and sensitivity to lighting changes, and enhancing the precision of motion target recognition. The findings of this research provide an effective technical solution for monitoring and defending against potential threats, particularly in countering the encroachments of irregularly shaped UAV s, UGVs, and specialized robots.