MEC-Based Evacuation Planning Using Variance Fractal Dimension Trajectory for Speech Classification

Evacuation models are used in crisis scenarios to optimize the path for occupants to escape a dangerous situation. The behavior of occupants in crisis scenarios has been demonstrated to differ between genders. This paper proposes an end-to-end system that can automatically model an evacuation plan based on the distribution of genders in a given space. Using MEC connected edge devices that can detect speech signals, and interfacing with the cloud through edge servers equipped with learning capabilities, an evacuation model can be generated in real-time according to current gender distribution in an occupied area. The key to making this system successful is an accurate gender speech classifier computed at the edge level. The classification model used was an SVM classifier, and the features used were MFCCs and VFDTs. The MFCCs displayed a gender classification accuracy of 91.47%, and when adding VFDT features, the gender learning accuracy was increased to 92.19%. This shows the benefit of adding VFDT features to speech classification accuracy. Indeed, running the classifier on the edge level is enabling the system to meet the maximum delay deadline (10 sec), as required by the NFPA code standards.