2010 17th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE) Mexico City, Mexico November 11-13, 2020 Lane Line Detection Computer Vision System Applied to a Scale Autonomos Car: AutoModelCar Jos´ e ´ Angel Balbuena Palma Mechatronics Engineering Division Instituto Tecnol´ ogico Superior de Atlixco Puebla, M´ exico im140744@itsatlixco.edu.mx Mariana N. Ibarra Bonilla Mechatronics Engineering Division Instituto Tecnol´ ogico Superior de Atlixco Puebla, M´ exico mariana.ibarra@itsatlixco.edu.mx Ra´ ul Eusebio Grande Mechatronics Engineering Division Instituto Tecnol´ ogico Superior de Atlixco Puebla, M´ exico raul.eusebio@itsatlixco.edu.mx Abstract—This article introduces a vision system of an au- tonomous vehicle scaled to detect the lane lines in the Au- toModelCar category of robotics tournaments. The system is developed in C++ using a modular integration architecture in ROS (Robot Operative System) and OpenCV as a computer vision library, embedded in an NVIDIA Jetson Nano card. The system applies image processing, inverse perspective mapping, feature extraction, filtering and modeling of the lane lines by first and second order polynomial regression. The tests that validate the performance of the system are presented. Index Terms—vision, lane line detection, autonomous vehicle. I. I NTRODUCTION Currently, there is a technological race for the development of the autonomous car, which has captured the attention of the industry and researchers. Tesla’s Autopilot System [1], the Waymo cars [2], and the Spirit System for the DARPA Urban Challenge [3] are examples that technology will soon take control over from cars. A fundamental function in the autonomous driving of vehi- cles is the detection of the road and lanes, so current proposals combine multiple sensors, such as cameras, LIDARs (Light detection and Ranging), IMUs (Inertial Measurement Unit), GPS and employ vision techniques, mapping and artificial intelligence. Several approaches for automatic lane detection have been reported in the state of art [4], [5], [6], [7]. In [4] the authors apply image processing and feature point location detection to detect lanes. However, they report an execution time of 0.12 s (8.3 Hz), then it is advisable to reduce the processing time in real applications. In [5] the authors face the lane detection problem by a Neural Network with Ellipsoidal with Dendritic Processing. They test their proposal and present good results, however the elliptical filtering stage consumes 40% of the total processing time. In [6] the authors describe a method for Lane Recognition using Active Contours Model. In general, the method is based on image processing, Inverse Perspective Mapping and Active Contours Model technique adding a parameter related to features of road. They report good results by testing carried out with an autonomous model car, however they got some false positives points due to light reflecting in the road. In a more recent paper [7], the authors present a lane line detection algorithm based on histogram statistics and sliding window. The autors demonstrate the superiority of their method with respect to other proposals that use Hough transform or feature extraction, which report problems with the detection of curves. They show the performance of their algorithm on straight lane lines and curves. Due to the high costs and size of these systems, different proposals are also reported in the literature that validate the vision systems using scale models [8], [9], [10] and some present them in International competitions [11] and [12]. In Mexico, the Mexican Robotics Tournament presents the AutoModelCars category, which has its antecedent in the project “Visions of Urban Mobility ” of the dual year Mexico-Germany 2016-2017, promoted by Dr. Ra´ ul Rojas to promote the training of human resources and research in the area of autonomous vehicles [13]. In this competition, the autonomous driving capabilities of a model vehicle are tested within a controlled environment. The vehicles presented in this type of competition mainly use cameras, mounted on the vehicle, to detect the lines and intersection of the track and apply some steps in common: image processing, extraction of characteristics and modeling of the lane lines [14], [15] and [16]. This work presents a vision system to detect the lane line for the AutoModelCar category in robotics tournaments. The proposed system uses a modular integration system in ROS (Robot Operative System) and OpenCV as a computational vision library. The vision system is programmed in C++ language, embedded in a Jetson Nano NVIDIA card, which runs a GNU/Linux distribution, Ubuntu 18.04. The artificial vision algorithm is based on the proposal [4], using Inverse Perspective Mapping, histogram statistics and sliding window for the lane lines detection, but other image processing oper- ations are added to improve the performance of line detection by sliding window, such as the white areas filter. Finally the lane lines are modeled using polynomial regression. II. HARDWARE DESCRIPTION The platform consists of a scale model car-based chassis RC Car WLtoys 12428 4WD, a 15 kg-cm servomotor and 978-1-7281-8987-1/20/$31.00 ©2020 IEEE