Design and implementation of a model predictive controller for lane detection and tracking system

Lane detection and tracking are the key advanced features of an advanced driver assistance system (ADAS) or automated vehicles. Lane detection is the process of detecting white lines on roads, and tracking helps the vehicle stay on the desired path. It controls the motion model by using previously detected lane markers. The overall aim of this project is to develop a robust and reliable lane detection and tracking approach for complex road geometry (e.g., Clothoid form roads) for a variety of occlusion (e.g., rain, daytime vs. night-time) by implementing model predictive controller (MPC). The contributions of this thesis are three folds.  Firstly, the study comprehensively reviews the previous literature on lane detection and tracking for ADAS and identify gaps in knowledge for future research. There are limited studies that provide state-of-art lane detection and tracking algorithms for ADAS. The study compares different lane detection and tracking algorithms and analyses different datasets used to verify the algorithms and metrics used to evaluate the algorithms. Specifically, the review identifies and classifies the existing lane detection and tracking algorithms under three themes: feature-based, learning-based and model-based, which provide a comprehensive approach towards understanding the key characteristics of lane detection and tracking algorithms in the literature. Some patented works by vehicle manufacturers under these three categories are also reviewed to acknowledge growing commercial interest in this field. This comprehensive review is expected to assist researchers working in this area by delivering current advancements made in lane detection and tracking for ADAS and the challenges to overcome in the future for robust lane detection and tracking systems. Secondly, this study proposes a framework for an innovative interpolation approach for quickly generating reliable ground truth annotation for lane detection and tracking system. Researchers are constrained to test their lane detection algorithms because of the small publicly available datasets. Additionally, those datasets may not represent differences in road geometries, lane marking and other details unique to a particular geographic location. Existing methods to develop the ground truth data sets are time intensive. To address this gap, this study proposes a framework for an interpolation approach for quickly generating reliable ground truth data. The proposed method leverages the advantages of the existing manual and time-slice approaches. Synthetic data has been collected using a monocular camera mounted at the front of a vehicle with a speed limit range of 60–80 km/h. A detailed framework for the interpolation approach is presented, and the performance of this approach is compared with existing methods. Video datasets for performance evaluation were collected in Melbourne, Australia. Results show that the proposed approach outperformed four existing approaches with a reduction in time for generating ground truth data in the range of 4.8% to 87.4%. A reliable and quick method for generating ground truth data, as proposed in this study, will be valuable to researchers as they can use it to test and evaluate their lane detection and tracking algorithms. Thirdly, and finally, the thesis develops a mathematical and simulation model for lane detection and tracking systems for clothoid-shaped roads to avoid occlusion in challenging weather conditions (e.g., rainy weather). The existing models have variety of occlusion issues for complex road geometry, which is greatly concerning for passenger safety. The proposed model in this thesis overcome those challenges. A lane detection and tracking algorithm is a clever method to decrease driving duties by eliminating the need for the driver to steer. The most fundamental prerequisite for an algorithm such as this is a mathematically well-developed route or trajectory. The first step relates to trajectory generation from the equations of clothoid and straight roads. This study designs a trajectory for an autonomous vehicle to study the fundamentals of the motion of a vehicle on a clothoid road. This system requires a Four-Wheel-Steering (FWS) vehicle to operate. This road designed from the equations is used as a sample road to find the steering angle, yaw rate and sideslip of the vehicle. The next step is to design a strategy to keep the vehicle on the desired path to avoid occlusion (larger distance error to the path) while travelling on clothoid roads. By designing the MPC (model predictive controller)-PC (preview capability) strategy to adjust the steering angle, this strategy controls the dynamic turning centre of the vehicle in such a way that the vehicle angle coincides with the turning centre of the road at any place. This method operates by shifting the kinematic centre of rotation to correspond with the actual vehicle centre of rotation and the road centre of curvature. The steering angle and kinematic centre of rotation must be adjusted to keep the vehicle on track. In the last step, a vehicle dynamic model is developed to find the real and resulting path of motion (cross-track error). The mathematical model developed is based on the differential equation of motion that calculates the kinematic steering angles needed to keep the vehicle in the desired direction of motion. The developed model determines the actual direction of motion by considering the vehicle’s sideslip when calculating the direction of motion. This is achieved by solving the vehicle’s steady-state dynamic equation for the desired direction of motion. Some techniques, such as feature-based and model-based models, have been used for lane detection and tracking on straight structured roads, but the learning-based approach has yet to be integrated with the MPC for structured and unstructured roads. Based on this study, it is found that MPC (Learning-based approach) is a better choice to avoid false detection. The proposed algorithm uses the steering angle, yaw angle and sideslip angle as inputs for the adaptive controller. MPC was used for planning difficulties in the predictive nature of the optimisation and the internal plant model used in the process. Instead of directly employing the optimisation problem solution for low-level control, MPC was applied to the internal plant model to provide reference output trajectories for another controller to track. The parameters of an MPC should be properly created and modified to provide good performance without causing discomfort. The MPC parameter contains the prediction horizon, the control horizon and the performance index weights. We demonstrate the parameter design and tuning method that was gained from multiple closed-loop simulations and experiments. Simulation tests for the lane detection method were carried out by analysing a road driving video in Melbourne, Australia (primary annotated ground truth data from interpolation approach) and the BDD100K datasets created by the Berkeley DeepDrive and other publicly available datasets such as the Industrial Consortium, KITTI, CULane and Caltech. The performance of the developed model is evaluated by using an intrusion detection metric used in the literature because it is more reliable. The simulation model is developed and tested in MATLAB with synthetic datasets. The mean detection accuracy ranges from 97% to 99%, and the detection time ranges from 20 to 22 ms under various road conditions with our proposed algorithm. This lane detection algorithm outperformed conventional techniques in terms of accuracy and processing time, efficiency in lane detection and overcoming road interferences. The proposed algorithm will contribute to advancing the lane detection and tracking of intelligent vehicle driving assistance and help further improve intelligent vehicle driving safety.