Intelligent sniper localization technique using convolutional neural network

Abstract

A reliable methodology for sniper localization utilizing a machine-learning framework has been proposed herein. The proposed work is based on the time of arrival of the shock waves (SW) in contrast to the conventional approaches, which utilized both the muzzle blast (MB) and SW. Since the MB is susceptive to environmental disturbances, the proposed solution is robust and reliable. The SWs are captured with the 2D non-linear array, and the time delay between the microphones is approximated using the generalized cross-correlation phase transfer (GCC-PHAT). Subsequently, a convolutional neural network (CNN) model is trained to map the input GCC-PHAT features to the sniper position. Adopting the CNN model provides robustness in the method, which also performs better in highly noisy environments. The performance of the proposed technique shows a significant improvement compared to the conventional methods providing a motivation to be used in practical applications.