An Unsupervised TinyML Approach Applied for Pavement Anomalies Detection Under the Internet of Intelligent Vehicles

Vehicles have been endowed with new technologies in the last years, mostly influenced by an increasing in the instrumentation level and availability of smart devices for embedded processing. In this scenario, which has paved the way for the construction of Internet of Intelligent Vehicles, the edge computing paradigm emerges with the primary role to promote the processing of raw data streams in their early stages, as close as possible to their origins. Therefore, this paper proposes the processing of data streams based on an unsupervised tiny machine learning approach to detect anomalies on the roads, exploiting for that a microcontroller and an embedded accelerometer on vehicles. The obtained results through real experiments were promising: the $f_{1}$ score mean was 0.76 for the first driver and 0.78 for the second. This indicates that the classifier model reached significant performance in the defined scenarios.