conclusion
The focus of this senior design report was to localize a sniper using a network of acoustic sensors based on the muzzle blast model. The maximum likelihood method was utilized to estimate the shooter position, which relied on TDOA measurements from multiple single-sensor nodes.
This system was modeled in Matlab, for six sensors and one sniper. Accuracy of the simulation was examined by varying the value of the standard deviation of the noise, which followed a normal distribution. For the method, the Cramér-Rao Bound was derived to measure the position error. To examine localization performance, the sensor geometry was varied. The results showed a robust model, able to localize a sniper for different sensor networks.
The concept of detection theory was applied to the model for the scalar case of two sensors. Detection investigation generated an accurate ROC curve, indicating a strong test.
Further research on this topic could begin by varying different parts of the model: changing the number of snipers, or the sensor network geometry. To enhance the model for a more realistic setting, adding walls or other obstacles could be useful. In addition to the acoustic sensor, augmenting the system with an optical sensor could provide additional data. The model in this report was based on detecting the muzzle blast of the gunshot. Continued research could simulate the system for the shockwave model, or the combined muzzle blast and shockwave model.
This report could be greatly enhanced by comparing the simulated results to real shooter data collected by acoustic sensors. The authors of this report contacted multiple sources requesting this data, including James Beldock, CEO of ShotSpotter, Inc., and David Lindgren, author of “Shooter localization in wireless microphone networks,” yet received no reply.
This system was modeled in Matlab, for six sensors and one sniper. Accuracy of the simulation was examined by varying the value of the standard deviation of the noise, which followed a normal distribution. For the method, the Cramér-Rao Bound was derived to measure the position error. To examine localization performance, the sensor geometry was varied. The results showed a robust model, able to localize a sniper for different sensor networks.
The concept of detection theory was applied to the model for the scalar case of two sensors. Detection investigation generated an accurate ROC curve, indicating a strong test.
Further research on this topic could begin by varying different parts of the model: changing the number of snipers, or the sensor network geometry. To enhance the model for a more realistic setting, adding walls or other obstacles could be useful. In addition to the acoustic sensor, augmenting the system with an optical sensor could provide additional data. The model in this report was based on detecting the muzzle blast of the gunshot. Continued research could simulate the system for the shockwave model, or the combined muzzle blast and shockwave model.
This report could be greatly enhanced by comparing the simulated results to real shooter data collected by acoustic sensors. The authors of this report contacted multiple sources requesting this data, including James Beldock, CEO of ShotSpotter, Inc., and David Lindgren, author of “Shooter localization in wireless microphone networks,” yet received no reply.