Road Localization in GrabMaps

6 posts · Feb 2020 – Nov 2023

What this trail covers

GrabMaps has evolved its map data processing by integrating hyperlocal data and leveraging big data solutions to improve map accuracy. This includes processing OpenStreetMap data, converting it to Parquet, and identifying hyperlocal attributes like road direction and speed limits. The system enhances map data with inputs from ride/delivery GPS pings and KartaView imagery, leading to improved accuracy of turn-by-turn navigation (GrabNav) through features like road localization, language, driving.

All Grab trails

Nov 2023Road localisation in GrabMapsThis post details the implementation of a road localization system for GrabMaps. The core technical contribution is the use of geohashes as a proxy to approximate geographical areas and localize road segments, significantly reducing computational cost compared to direct geometric inclusion tests. The process involves precomputing geohashes for areas and then assigning roads to these geohashes. The post also discusses precision tradeoffs, exploring solutions like increasing geohash precision, calculating geohash coverage percentage, and using border shapes for geohashes near boundaries to improve accuracy.
Aug 2023Building hyperlocal GrabMapsThis post details the process of integrating hyperlocal data into GrabMaps, focusing on road network attributes. It describes downloading and converting OpenStreetMap .pbf files to Parquet format using a Parquetizer. The identification of hyperlocal data involves applying customized rules defined by local map teams, combined with data from GPS pings and KartaView. A specific example of intersection link identification is provided, explaining the use of Spark, geohash, and graph computations to determine when road segments meet specific criteria (parallel, same name, one-way, angle/shape). The impact is demonstrated through improved routing efficiency in GrabNav.
Feb 2023New zoom freezing feature for Geohash pluginIntroduced a 'zoom freeze' feature to the Geohash JOSM plugin. This feature allows users to toggle the display of new geohashes when zooming in or out, improving map readability and user experience. The implementation involves a UI label that changes from 'Enable zoom freeze' to 'Disable zoom freeze' and disables options for displaying larger/smaller geohashes when enabled.
Dec 2022How KartaCam powers GrabMapsThis post introduces KartaCam, an AI-powered camera designed for efficient and low-cost map-making. KartaCam features a wide field of vision, high image quality, edge AI for smart capturing (scene recognition, image quality checking, object detection, privacy blurring), a dual-band GNSS module for improved positioning accuracy, and a 4G LTE module for remote management. KartaCam 360, using four synchronized cameras, captures panoramic views. The KartaCam system is integrated into the GrabMaps workflow, which includes collection management tools, the KartaView mobile app, and automatic processing for quality checks and PII blurring. This initiative aims to reduce operational costs, improve data quality and map attributes, and uphold privacy standards by blurring sensitive information at the edge.
Jul 2021Protecting Personal Data in Grab's ImageryThis post details the development of KartaView, Grab Geo's platform for geotagged imagery, and the subsequent implementation of an image blurring process to protect personal data. The blurring process involves transforming images into planar formats, using an object detector (YOLOv4) to identify and blur faces and license plates, and addressing challenges with diverse image types and diverse license plate designs. The dataset evolved through multiple iterations to handle scenarios like masked faces, small or rotated objects, and variations in license plate designs. Data augmentation techniques, including offline (image view splitting, oversampling) and online methods, were employed to improve model performance. The post also touches upon the integration of KartaView imagery with GrabMaps for improved road localization and navigation.
Advertisement
Feb 2020Grab-Posisi - Southeast Asia’s First Comprehensive GPS Trajectory DatasetThis post introduces Grab-Posisi, Southeast Asia's first comprehensive GPS trajectory dataset. It details the dataset's collection from Grab drivers' phones, its rich contextual information (accuracy, bearing, speed) with a 1-second sampling rate, and its extensive coverage in Singapore and Jakarta. The post outlines applications such as on-map inference, map matching, traffic detection and forecast, mode detection, and economic analysis for urban planning. It also provides instructions for accessing the dataset for research purposes and cites the relevant publication.