Global visual localization in LiDAR-maps through shared 2D-3D embedding space
This work addresses the under-explored challenge of matching images to LiDAR-maps for robotic applications like autonomous driving, offering a novel solution that could enhance localization in high-definition 3D environments.
The paper tackles the problem of global visual localization within LiDAR-maps by developing a deep neural network approach that creates a shared embedding space between images and 3D point clouds, enabling image-to-3D-LiDAR place recognition. The method was evaluated on the Oxford Robotcar Dataset under various weather and light conditions, showing effectiveness across different learning paradigms, architectures, and loss functions, though specific numerical results are not provided in the abstract.
Global localization is an important and widely studied problem for many robotic applications. Place recognition approaches can be exploited to solve this task, e.g., in the autonomous driving field. While most vision-based approaches match an image w.r.t. an image database, global visual localization within LiDAR-maps remains fairly unexplored, even though the path toward high definition 3D maps, produced mainly from LiDARs, is clear. In this work we leverage Deep Neural Network (DNN) approaches to create a shared embedding space between images and LiDAR-maps, allowing for image to 3D-LiDAR place recognition. We trained a 2D and a 3D DNN that create embeddings, respectively from images and from point clouds, that are close to each other whether they refer to the same place. An extensive experimental activity is presented to assess the effectiveness of the approach w.r.t. different learning paradigms, network architectures, and loss functions. All the evaluations have been performed using the Oxford Robotcar Dataset, which encompasses a wide range of weather and light conditions.