Real-Time lane departure warning system on an embedded platform

In this work, a real-Time lane departure warning system with its embedded implementation is proposed. In the proposed method, unintended lane departures that are originating from drivers falling asleep, carelessness etc. are detected using image processing based approaches. In this study, initially lane markings are extracted by making use of a filter to provide lane marking detection process robust against to possible outliers. At the next stage, correlation of the input image segments with a 1-D Gaussian function is computed to determine candidate lane markings. Possible outliers at this stage are eliminated using RANSAC (RANdom SAmple Consensus) approach and lane markings are obtained. Additionally, temporal relationship of the detected lane markings is examined via Kalman filter. Vehicle lane departure decision is given by interpreting the obtained lane markings and predetermined vehicle positions. The developed system is able to process 16 fps on a 1 GHz ARM Cortex A8 processor for input images of size 752×480.