Fall detection through acoustic Local Ternary Patterns

In this paper, we propose a framework that detects falls by using acoustic Local Ternary Patterns (acoustic-LTPs) by analyzing environmental sounds. The proposed method suppresses silence zones in sound signals and distinguishes overlapping sounds. Acoustic features are extracted from the Separated source components by using the proposed acoustic-LTPs. Subsequently, fall events are detected through a support vector machine (SVM) based classifier. The performance of the proposed descriptor is evaluated against state-of-the-art methods that are applied on well-known sound databases. A comparative analysis demonstrates that the proposed descriptor is more powerful and reliable in terms of fall detection than other methods, and it also performs well in a multi-class environment. Moreover, the proposed descriptor possesses a rotation invariant property, and therefore, it demonstrates significant resistance against the rotated sound signals.