"The proposed method first uses the feature pyramid network (FPN) is used to detect the humans from the aerial image. Next, the proposed Scatter Net Hybrid Deep Learning (SHDL) network is used to estimate the pose for each detected human."

"Finally, the orientations between the limbs of the estimated pose are used by the support vector machine (SVM) to identify individuals engaged in violent activities," stated the paper, explaining its methodology. The drones used were standard Parrot AR 2.0. For now, the work remains unstable with the accuracy of detection significantly decreasing the bigger the number of violent individuals in a test group gets. A crowd of 10 with only one violent offender results in an acceptable accuracy of detection of over 94%.However, raising that number to just two violent individuals sees the accuracy drop to 90.6%, while raising it to five sees disappointing results of 84%. The researchers say this malfunction may be due to the "inability of the FPN network to locate all the humans or the incapability of the SHDL network to estimate the pose of the humans accurately."