Real time lane detection and tracking system evaluated in a hardware-in-the-loop simulator

This paper presents a new real time lane detection and tracking system which succeeds in sharp curved roads. The Warp Perspective Mapping (WPM) method was used to change the video image perspective to top view. A two dimensional Gabor filter and vertical pixel density calculation method was used to find initial lane regions. Second derivative approach of Composite Bezier spline fitting method calculates the exact coordinates of the control points. Score function based on pixel matching was used for determining lanes in curved roads. A template lane matching method with ease of implementation was used for tracking the detected lanes. The proposed system was tested using offline and real time hardware-in-the-loop simulations. The detected and tracked lane information is used in a lane keeping control system, realized in a hardware-in-the-loop setting. The high fidelity, high order, realistic and nonlinear vehicle model in Carmaker HIL is used in hardware-in-the-loop testing. The lane keeping controller works as code in a dSpace microautobox general purpose electronic control unit which obtains the lane data from the vision system running on a laptop PC and Carmaker vehicle data from a dSpace compact simulator for calculating the required steering actions. The system performance was evaluated on different scenarios which contain straight and sharp curved roads, obstacle vehicles and different lane marking types. Average success rate is calculated as % 98.7.