Threshold Based Algorithm For Bicycling Crash Detection Empirical Evaluation and Optimization / Ak Mohammad Salihin Pg Hj Abd Rahim

A Dissertation Submitted in Partial Fulfillment for the Degree of Msc By Coursework i Computing and Information systems Universiti Teknologi Brunei.

Abstract
Despite the growing popularity of bicycles as an alternative to motor vehicles, the world has categorized bicycles as one of the vulnerable road users on the road. The widespread use and increase in the capability of smartphones serve as a motivation for developing more cost-effective solutions to boost bicycling safety. The purpose of this research is to support further the needs for enhancing the safety of the bicyclists by proposing a solution that combines the ability of a smartphone and a threshold-based algorithm (TBA) for bicycling crash detection. Multiple studies have concluded that TBA is effective in detecting fall incidents. However, the majority of the studies on TBA focused more on detecting fall amongst the elderly, and there is inadequate research on the use of this algorithm for a bicycling crash. In addition to this, the findings from the studies on TBA also reported that the algorithm has a limitation of a high false positive rate, and it requires some amendments in its structure to optimize its performance. Therefore, this research aimed to contribute new findings by applying the TBA to the bicycling domain and suggest ways to optimize the algorithm, so it fits the researched domain best. The optimization process involved developing a mobile application prototype entitled Cyclists Fall to generate the dataset for finding a reliable threshold value to evaluate the algorithm. The findings presented underlined that the application of the basic TBA in the bicycling domain has similar behavior as the elderly fall, where it performs very well in detecting bicycling crashes, but the false positive rate remains high. Due to this reason, this research considers two optimization methods to improve the algorithm, the Acceleration-Drop method and the Multiple-Exceeds method. The first method was rejected due to its high implementation complexity. On the contrary, the second method shows that it has effectively reduced the false positive rate in the result, and the precision has increased significantly from 62.5 per cent to 84 per cent.

Dissertation (Degree of Msc in Computing and Information Systems)

Includes bibliography references.