Copula Based Moving Vehicle Classification in Compressive Domain without Signal Reconstruction

Economical, optimal and accurate solutions for vehicle/personnel detection is becoming increasingly important especially in the defense domain, border surveillance and other urban infrastructural operations. It is important that only a minimal number of data samples are used to arrive at an accurate decision statistic due to the physical and computational constraints faced in most of these applications. In this paper, two sensor modalities namely, acoustic sensors and seismic sensors are used to collect data samples. We explore two different approaches for the problem at hand. Firstly, we use compressive sensing to filter out only the required number of samples. For this, we have created a novel algorithm for building the basis matrix used as part of CS. Secondly, we leverage the data dependencies from the two sensor modalities by using data fusion. We use copula functions to fuse data samples together. Using both the mentioned techniques the novel algorithm creates a reliable decision statistic that can be used for detection. As a result there is no need for the initial signal to be reconstructed. Finally, the paper also analyses and compares detection accuracies obtained as a result of various basis matrices and copula functions. Simulations show that our algorithm improves the detection speed significantly, reducing the computation load and providing with a detection accuracy of 92.68% without the need of signal reconstruction.