2019/10/02 by Daniele Cattaneo, Matteo Vaghi, Cattaneo, Daniele +7 · 1 citation
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Robotics (cs.RO) #Robotics and Sensor-Based Localization #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1910.04871
openalex publication_date 2019/10/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Global localization is an important and widely studied problem for many\nrobotic applications. Place recognition approaches can be exploited to solve\nthis task, e.g., in the autonomous driving field. While most vision-based\napproaches match an image w.r.t. an image database, global visual localization\nwithin LiDAR-maps remains fairly unexplored, even though the path toward high\ndefinition 3D maps, produced mainly from LiDARs, is clear. In this work we\nleverage Deep Neural Network (DNN) approaches to create a shared embedding\nspace between images and LiDAR-maps, allowing for image to 3D-LiDAR place\nrecognition. We trained a 2D and a 3D DNN that create embeddings, respectively\nfrom images and from point clouds, that are close to each other whether they\nrefer to the same place. An extensive experimental activity is presented to\nassess the effectiveness of the approach w.r.t. different learning paradigms,\nnetwork architectures, and loss functions. All the evaluations have been\nperformed using the Oxford Robotcar Dataset, which encompasses a wide range of\nweather and light conditions.\n