Smartphone-Based Collision Alert System for Motorcyclists

The primary aim of this study is to develop a FCW system tailored for motorcyclists. This system is designed to identify potential collisions by analyzing the time-to-collision and trajectories of both the detected vehicles and the motorcycle itself. The development encompasses three distinct approaches. Firstly, the time-to-collision metric is derived using input from a cost-effective camera. Secondly, the trajectory of detected vehicles is forecasted based on video data represented in 2D pixel coordinates. Thirdly, the lean direction of the motorcycle, vital for trajectory prediction, is determined using data from a low-cost inertial measurement unit sensor.The culmination of these approaches results in an integrated Advanced FCW system. For the prediction of time-to-collision, a nested Kalman filter is employed in conjunction with vehicle detection techniques, converting image pixel data into relative distance, velocity, and time-to-collision information. To forecast the trajectories of detected vehicles, various algorithms were evaluated, with LSTM demonstrating superior performance on the dataset. Lastly, it was determined that measuring the lean angle of the ego vehicle is more effective for determining its leaning direction than employing riding pattern classification methods.