Matching Cross-Domain Data with Cooperative Training of Triplet Networks: A Case Study on Underwater Robotics

Giovanni G. De Giacomo, Matheus Machado dos Santos, Paulo L. J. Drews-Jr, Silvia S. C. Botelho

Abstract: Recently, Deep Convolutional Neural Networks have been successfully applied to various robotics problems, such as robot vision and simultaneous localization and mapping. Among these, siamese and triplet networks have obtained great traction in intra-domain matching. However, it is impossible to directly use these networks in cross-domain problems. Thus, this paper proposes a new method to train a set of triplet networks to perform cross-domain matching and ranking focused in underwater robotics. The method is used to train a pair of networks to perform matching of acoustic and segmented aerial images aiming to support an underwater robot localization algorithm. Results show that the method is able to achieve up to 83% accuracy in matching acoustic and segmented aerial images and up to 85% recall in ranking relevant aerial images given an acoustic image.