Graph-Based Point Tracker for 3D Object Tracking in Point Clouds

In this paper, a new deep learning network named as graph-based point tracker (GPT) is proposed for 3D object tracking in clouds. GPT not based on Siamese applied to template and search area, but it the transfer of target clue from area. end-to-end trainable. has two modules: graph feature augmentation (GFA) improved (ITC) module. The key idea GFA exploit one-to-many relationship between area points using bipartite graph. GFA, edge features are generated by transferring clues through convolution. It captures effectively perspective geometry shape second module ITC. ITC embed information center into edges via Hough voting, strengthening discriminative power GFA. Both modules significantly contribute improvement geometric including effectively. Experiments KITTI dataset show that achieves state-of-the-art performance can run real-time.