DEVELOPMENT OF LOCATION DETERMINATION SYSTEM BY INTEGRATING MAP MATCHING ALGORITHM AND VISION-BASED SYSTEM IN AUTONOMOUS VEHICLES

• Main Author: Rahma Jumardi, Aulia
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84142
• Institution: Institut Teknologi Bandung

Autonomous vehicles are vehicles that operate by minimizing manual human intervention to move independently towards a predetermined destination. Therefore, autonomous vehicles require an accurate localization system. The commonly used localization system is GPS, but only using GPS for the localization system is not optimal enough. The accuracy of GPS often decreases because the use of GPS is very vulnerable to the surrounding environment such as bad weather, tunnels, buildings, trees and low sampling rates. To overcome this, map-based localization using digital maps as a reference for vehicle positioning is integrated with GPS so that it can be a solution to improve GPS accuracy. The method used is called map matching algorithm. The map matching algorithm is a method for integrating digital map data with data from positioning systems such as GPS to identify the path traveled by the vehicle and determine the exact position of the vehicle on that path. Map matching algorithms are used to correct and improve positions detected by GPS or other localization systems that may be less accurate. This research uses three map matching algorithms namely Iterative Closest Point (ICP), Spatial Map Matching (ST-Matching) and Hidden Markov Model (HMM). These three methods can be used with only two inputs, namely trajectory information from GPS and reference map coordinate information that can be obtained from Google Earth. To improve the accuracy of the map matching results, this research proposes a vision-based system using camera sensors, especially for roads that have at least two lanes for vehicles to pass. The camera detects road markings so that information is obtained on which lane the vehicle is running. Based on this local locator system information that is used as input to determine the lane reference map used then the map matching algorithm can be applied. The resulting vehicle position is not only the position of the vehicle on the road but more specifically in the lane the vehicle is traveling in. This study was conducted using real data taken at two different locations. The first location is a straight road without any intersections. The results showed that the performance of the ICP algorithm has the smallest computational weight of 608501 and the fastest execution speed of 1.09 seconds compared to the other two algorithms. For the accuracy part, the first location has a total number of GPS points of 59 points with the highest accuracy weight of 118. The ICP and ST-matching algorithms only have 3 positioning errors, namely at points 15, 33 and 37. Then the percentage of accuracy of ICP and ST-matching is 94.91%. While the HMM algorithm has the closest point as many as 29 points, still within the local search map limit of 27 points and a positioning error of 3 points. Then the percentage of accuracy of the HMM algorithm is 72.03%. This proves that the ICP algorithm has the best performance on straight roads. The second location is a road that has two intersections. The results show the performance of the HMM algorithm has the smallest computational weight of 192026 with an execution speed of 1.17 seconds while the ICP algorithm has the fastest execution time of 0.56 seconds, and the ST-matching algorithm has the largest computational weight of 297201. For the accuracy part, the second location has a total number of GPS points of 17 points. Because this second location has 2 intersections, however, the intersection used is only the first intersection, the total accuracy weight of the second location is 36. The ICP and ST-matching algorithms have 2 positioning errors, namely at the 4th point and positioning errors at the intersection. Then the percentage of accuracy of ICP and ST-matching becomes 72.22%. While the HMM algorithm has the closest point as many as 13 points, still within the boundaries of the local search map as many as 2 points, 1 point positioning error and right in overcoming intersection uncertainty. Then the percentage of accuracy of the HMM algorithm is 88.88%. This proves that the HMM algorithm has the best performance on roads that have intersections. Keywords: Autonomous Vehicle Localization, GPS, Reference Map, Map Matching, Camera. ?