Abstract
As technology advances, the demand for autonomous cars and Advance Driver Assistance System (ADAS) is on the rise and this include a wide range of functions such as Forward Collision Warning (FCW), Lane Departure Warning, Blind Spot Detection, Automatic Emergency Braking. Currently, most light weight Forward Collision Warning system predicts if a car is a threat if it detects a car within a certain threshold such as displacement, velocity and even time to collision. This, however, fails to take into the account of the dynamics of the road situation as a vehicle detected which is close to the ego vehicle is not necessarily a threat if it is on a different lane. Hence, this thesis proposes an advance Forward Collision Warning system which detects potential colliion by determining if the detected vehicle will enter the ego vehicle travel path through trajectory prediction. The proposed method can be divided into three categories, namely: 1) forward collision warning system using nested Kalman filter on a monocular camera; 2) trajectory prediction of vehicle coordinates using Long Short Term Memory (LSTM) and 3) leaning direction prediction based on IMU sensor.