Road tracking using particle filters for Advanced Driver Assistance Systems

Road segmentation and tracking is of prime importance in Advanced Driver Assistance Systems (ADAS) to either assist autonomous navigation or provide useful information to drivers operating semi-autonomous vehicles. The work reported herein describes a novel algorithm based on particle filters for segmenting and tracking the edges of roads in real world scenarios. This is accomplished with the help of a video camera mounted on the vehicle. The measurement and prediction functions in particle filtering are modified suitably to measure and track road edges with time. One road measurement function and two different prediction functions are proposed and their performances are compared. The proposed measurement function is based on the familiar K-means clustering algorithm. The two prediction models considered are polynomial curve fitting and first order Auto Regressive (AR) models. This method of segmenting and tracking roads is tested on over 40 videos obtained from vehicles moving over a range of real world highways. The results demonstrate that the proposed method is capable of handling partial occlusion of the road due to the presence of multiple objects, texture variations in the road, and abrupt illumination changes, such as extensive shadowing. Also owing to the fact that this approach is capable of operating satisfactorily at resolutions as low as 160×120 pixels, it places a very minimal computational load on the onboard Electronic Control Unit (ECU). By virtue of its robustness and ability to continuously adapt to changes in road conditions, the proposed method is a valid candidate for real time implementation.