An efficient moving object tracking framework for WSNs using sequence-to-sequence learning model

Wireless sensors can detect an object from the light it reflects, the noise it causes, or the gas molecules it disseminates. However, tracking a moving object requires the wireless sensors to perform high-frequency sensing and data transmission which consume much more energy. To save energy and prolong the lifetime of wireless sensor networks while tracking a moving object effectively, this paper proposes a framework that predicts the trajectory of the moving object using a Sequence-to-Sequence learning (Seq2Seq) model and only wakes-up the sensors that fall within the predicted trajectory of the moving object with a specially designed control packet. The framework uses DV-Hop (distance vector of hops to anchors) as the virtual coordinate that eliminates the dependency of using GPS to locate the sensors to be invoked for tracking the moving object. The framework translates the object’s moving trajectory to a sequence of cascaded hyperbolas and encodes the hyperbolas with DV-Hop constraints. A control packet containing these constraints forbid sensors not in the trajectory to rebroadcast, and awake/sleep signals that control the sensors’ action. The proposed Seq2Seq model predicts the target’s next trajectory directly and outputs a control message that could route along the predicted trajectory. In comparison to predicting the target’s trajectory then encoding the trajectory using geometric objects such as hyperbola, the proposed Seq2Seq model reduces the computation time of encoding geospatial trajectory. Also, the proposed framework preserves the location anonymity by only transmitting the hop’s information instead of GPS values. The performance comparisons with the existing methods show an improvement in energy-saving and control message routing delay.