Learning geometric soft constraints for multi-view instance matching across street-level panoramas

Abstract

We present a new approach for matching tree instances across multiple street-view panorama images for the ultimate goal of city-scale street-tree mapping with high positioning accuracy. What makes this task challenging is the strong change in view-point, different lighting conditions, high similarity of neighboring trees, and variability in scale. We propose to turn (tree) instance matching into a learning task, where image-appearance and geometric relationships between views fruitfully interact. Our approach constructs a Siamese convolutional neural network that learns to match two views of the same tree given many candidate tree image cut-outs and geographic information of the two panorama images. In addition to image features, we propose utilizing location information about the camera and the tree. Our method is compared to existing patch matching methods to prove its edge over state-of-the-art. This takes us one step closer to the ultimate goal of city-wide tree mapping based solely on panorama imagery to benefit city administration.