Attention-SLAM: A Visual Monocular SLAM Learning From Human Gaze

Abstract

This paper proposes a novel simultaneous localization and mapping (SLAM) approach, namely Attention-SLAM, which simulates human navigation mode by combining a visual saliency model (SalNavNet) with traditional monocular visual SLAM. Firstly a visual saliency model namely SalNavNet is proposed. In SalNavNet, we introduce a correlation module and propose an adaptive Exponential Moving Average (EMA) module. These modules mitigate the center bias, which most current saliency models have. This novel idea enables the saliency maps generated by SalNavNet to pay more attention to the same salient object. An open-source saliency SLAM dataset namely Salient-Euroc is published, it consists of Euroc dataset and corresponding saliency maps. Moreover, we propose a new optimization method called Weighted Bundle Adjustment (Weighted BA) in Attention-SLAM. Most SLAM methods treat all the features extracted from the images as equal importance during the optimization process. In weighted BA, the feature points extracted from the salient regions have greater importance. Comprehensive test results prove that Attention-SLAM outperforms benchmarks such as Direct Sparse Odometry (DSO), ORB-SLAM, and Salient DSO in the 7 of 11 test cases. The test cases are all indoor scenes, with varying brightness, speed, and image distortion. Compared with ORB-SLAM, our method improves the accuracy by 4% and efficiency by 6.5% on average.