Automatic gunshot detection technology allows incidence response system to counteract the potential of crimes. However, the surveillance systems suffer from various detection problems, such as difficulty in differentiating gunshot, fire work and other similar sounds. To improves the accuracy and reduces processing time, we have proposed hybrid algorithm for automatic detection of gunshots in indoor environment. In the proposed approach, we have used pre-processing steps which filters the input audio signals with a threshold. During pre-processing, the signals having smaller energy than the threshold value are discarded because these low energy signals are normal sound signals. When energy of audio signal is more than the threshold value and deemed ambiguous audio, such signal is forwarded to next step for further processing. The second step of the proposed approach is based on features based algorithm, in which antilog energy features are implemented to increase accuracy. These features extend energy band to easily differentiate between gunshot and normal scream. For classification purpose, SVM, Tree and KNN classifiers are used comparatively to differentiate a classifier which will show more accuracy with minimal computational cost. The proposed approach provides 94.97% accuracy for SVM,92.56% accuracy for KNN classifier, and 91.65% accuracy for Tree classifier. The pre-processing step reduces computational time by 5%, 13.61% and 34.56% for KNN, Tree and SVM classifiers respectively. The pre-processing step in the proposed algorithm requires 5.80% processing time of features based approach to filter an audio signal.

自动枪击检测技术使事件响应系统可以抵消潜在的犯罪行为。但是，监视系统遭受各种检测问题，例如难以区分枪声，射击工作和其他类似声音。为了提高准确性并减少处理时间，我们提出了一种混合算法来自动检测室内环境中的枪声。在提出的方法中，我们使用了预处理步骤，该步骤会过滤具有阈值的输入音频信号。在预处理期间，具有比阈值小的能量的信号被丢弃，因为这些低能量的信号是正常的声音信号。当音频信号的能量大于阈值并且被认为是模棱两可的音频时，该信号被转发到下一步以进行进一步处理。拟议方法的第二步基于基于特征的算法，其中实现了对数能量特征以提高准确性。这些功能扩展了能带，可以轻松区分枪声和正常尖叫声。出于分类目的，比较而言，使用SVM，树和KNN分类器来区分分类器，该分类器将以最小的计算成本显示出更高的准确性。该方法为SVM提供了94.97％的准确性，为KNN分类器提供了92.56％的准确性，为Tree分类器提供了91.65％的准确性。对于KNN，Tree和SVM分类器，预处理步骤分别减少了5％，13.61％和34.56％的计算时间。该算法的预处理步骤需要基于特征的方法对音频信号进行滤波的处理时间为5.80％。