11:00 - 12:00 | Mon 28 Oct | Crystal Room I | MoC-T5
A rapidly growing interest can be seen towards the use of ubiquitous sensing (e.g., smart-phones and in-vehicle sensors) for identification of road conditions and driver behaviors that adversely affect the comfort and safety of motorists. While such detection and classification tasks have achieved reasonable success, the overhead associated with big-data processing and classification possess challenges to low-cost low-bandwidth implementations. In this paper, by viewing the detection and classification problem as a discrete-time filtering task, a highly efficient approach that can be implemented in near real-time is presented. In particular, by focusing on vibration measurements (as generated by smart-phones or in-vehicle sensors), a bank of matched filters are developed for automatic detection and classification of several hazards, including identification of their type and time of car's impact with the hazard. Presented approach is illustrated using real-life vibration data. This technique provides sufficient conditioning that the data can be used to determine the time of impact, type, and location of each hazard without any manual input.
No information added