|
|
Place recognition method based on YOLOv3 and deep features |
1. School of Electronic Engineering, Changzhou College of Information Technology, Changzhou, Jiangsu 213164, China; 2. School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China; 3. School of Automation, Southeast University, Nanjing, Jiangsu 210096, China |
|
|
Abstract To eliminate the adverse effects of viewpoint and appearance on the application of place recognition, a method was proposed based on salient landmarks and deep features. The salient objects were extracted as candidate landmarks. The YOLOv3 network was designed to identify candidate landmarks for deleting specific object types not suitable for environmental modeling. In the image similarity measurement process, the convolutional neural network(CNN) features of candidate landmarks were extracted, and the dimension reduction operation was combined to improve the matching efficiency. The experiments were conducted on three challenging place recognition datasets. The results show that compared with the methods of FABMAP, SeqSLAM and PlaceCNN, the average accuracy of the proposed recognition method reaches 7122%, which is better than those of comparison algorithms.The CNN features on the filtered salient landmarks can adapt to the changes in viewpoint and appearance.
|
|
|
|
|
|
|
|