Smart Wireless Fence Based on Machine Learning and Coordinate Transformation

In recent years, with the increase in the number of high-end parks, the importance of blind spot security has become increasingly high, which means whether confidential points within the park can be protected for privacy. Some traditional methods, such as trajectory collection based on mobile signal base stations, are often costly, inefficient, and prone to blind spots. At the same time, trajectory prediction methods have low accuracy. Therefore, we propose an end-to-end wireless warning system, which includes multi-scale distributed visual sensor spatial transformation for trajectory collection, temporal convolution and Transformer based on dynamic time windows for trajectory prediction, as well as probability transfer and cosine trend based on predicted trajectory points for confidential point warning. Our experimental research shows that our proposed TCNFormer performs the best, providing novel ideas for future warning systems.