Research On Abnormal Sound Detection For Intelligent Surveillance In Public Places

With the development of the modern society, the environmental surveillance in public places, such as airports, banks, and military areas, attracts more and more attention. Because the abnormal sounds can reveal the information of accidents and critical affairs, the research on abnormal sound detection and recognition in public places is of great significance. Although some scientists and researchers have paid attention to sound detection in certain particular environments, as is well known that the research on abnormal sound detection in public places is very difficult and is relatively backward. We investigate and analyze lots of articles and papers about this research. Then based on the investigation, the detection and recognition of abnormal sounds in public places for intelligent surveillance are studied as following:①On account of that the abnormal sound databases used in researches are not uniform, we collect and pack up the public abnormal sounds, according to the occurring probability, harm extent and relative studies. Our own abnormal sound database is created, which includes the sounds of explosion, gun shots, scream, glass breaks, glass cracks and bottle cracks.②Based on the characteristic difference between the abnormal sounds and the environmental noise, the method of abnormal sound detection and recognition is studied. We detect the sound signals utilizing the variance of the signal energy by wavelet transform. According to the hearing mechanism, MFCC is extracted to characterize the abnormal sounds. The recognition system based on HMM is used to recognize the sound type. Experiments about the method are conducted on our own database.③Because the recognition rate is low in noisy environment when using MFCC to represent the abnormal sounds, the feature extraction is studied further to handle this issue. We analyze the spectral characteristics of the abnormal sounds and redesign the filter bank of MFCC to extract the improved parameters by using the obtained statistical characteristics. We conduct the experiment about the two parameters.④A real time surveillance system in public places is designed in details, based on abnormal sound detection and recognition. And we also develop the software of the module of the intelligent audio surveillance.