Evaluation of Gunshot Detection Algorithms

Abstract

Six preprocessing algorithms for the detection of firearm gunshots are statistically evaluated, using the receiver operating characteristic method as a previous feasibility metric for their implementation on a low-power VLSI circuit. Circuits are intended to serve as the input detection sensors of a low-power environmental surveillance network. Some possible VLSI implementations for the evaluated algorithms are also evaluated. Results indicate that the use of wavelet bank filters, either discrete or continuous, might be the best choice in terms of the compromise between detection efficiency and the power requirements of the intended application.

Effect of Sampling Rate Reduction and Signal Filtering for Gunfire Sound Classification with Spectral Vector using ANN

“…It was found that the least error in classification was obtained from the case of applying LS-LDA (Least Square-Linear Discriminant Analysis) to the system. To utilize VLSI with low power consumption in gunshot detection, ROC (Receiver Operating Characteristic) was adopted as the metric in measuring the performance [7]. It was shown that to get a good performance, Wavelet Transform has to be applied.…”

Improving efficiency and reliability of gunshot detection systems

“…The superior performance of this detection feature was then reported under noisy environments in comparison with other known methods of gunshot detection [4]. However, we have found in our experiments that even though the correlation method of [7] gives a high true positive rate (TPR), it yields an undesirably high false positive rate (FPR). As a means of reducing this FPR, we propose using cross-correlation maximum against a gunshot template in conjunction with eighth order linear predictive coding (LPC) coefficients [8] as features with a gaussian radial basis function (RBF) Kernel for a support vector machine (SVM) classifier [9].…”

“…The results of the aforementioned three experiments are shown in Table 1. Though correlation based template matching has been proposed as a high performance gunshot detection algorithm in [7] [4], it gives a high FPR with our database. Moreover, Figure 2 shows the inseparable overlap in cross-correlation maxima of gunshots and outsider signals with a gunshot template.…”

“…The gunshot detection problem has also been pursued as a maximization task employing a dynamic programming solution to the detection problem [3]. More recently, the use of correlation against a gunshot template has been proposed by Freire and Apolinario [7]. The superior performance of this detection feature was then reported under noisy environments in comparison with other known methods of gunshot detection [4].…”

On the Effectiveness of Dynamic Spectrum Management Algorithms in xDSL Networks

“…The problem of gunshot detection has been explored in the context of audio streams from movies [4] using dynamic programming and another work evaluates different algorithms in the task of detecting firearms gunshots [5]; the later reminds the reader that, being a gunshot signal similar to an impulsive signal, its spectral characteristics shall most likely provide information of the acoustic surroundings. The correlation detection algorithm has presented prospective good results [5]. This work evaluates the performance of a correlation measure against a template, by comparing it to more sophisticated methods such as HMM [6] working on LPC [7], MFCC [7], or the impulsivity parameter of stable distributions [8].…”