Bionic Binocular SLAM with Self-Attention Depth Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual SLAM systems face challenges in tracking failures due to sparse textures or dynamic objects, limiting their ability to accurately perceive and navigate unknown environments.
Innovation Solution
A bionic binocular active visual SLAM system is developed, utilizing a panoramic camera for peripheral vision and a binocular gaze control strategy, combined with a self-supervised depth estimation network based on a channel self-attention mechanism, to enhance scene understanding and navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a visual SLAM system passively receives image data during robot motion, then the system can perform localization and mapping, but the system experiences tracking failures when images have sparse textures or numerous dynamic objects
Solution Approach 1:
The system transitions from passive image capture to active gaze control, where the camera dynamically adjusts its viewing direction based on scene analysis. The gaze control module actively selects optimal viewing directions to capture images with rich textures and minimal dynamic objects, thereby maintaining reliable tracking while adapting to complex environments.
Solution Approach 2:
Before performing SLAM processing, the system performs preliminary scene analysis using a value map that evaluates scene characteristics (texture richness, dynamic object density). Based on this evaluation, the gaze control module pre-determines optimal viewing directions to capture high-quality images, preventing tracking failures before they occur.
2Measurement precision
If human eyes achieve rapid and precise gaze positioning and smooth tracking through vestibulo-ocular reflex, then visual perception capability is enhanced, but replicating this mechanism in robots requires complex bionic eye systems
Solution Approach 1:
The patent extracts only the essential gaze control function from the complex vestibulo-ocular reflex mechanism. Instead of replicating the entire biological system, it implements a simplified computational model that achieves precise gaze positioning through algorithmic control based on value map analysis, thereby reducing device complexity while maintaining measurement precision.
Solution Approach 2:
The patent replaces the mechanical biological mechanism (vestibulo-ocular reflex) with a computational system. The bionic eye control module uses image processing and algorithms to determine gaze direction, substituting the mechanical reflex arc with electronic computation to achieve similar functional outcomes with reduced complexity.
3Quantity of substance
If self-supervised depth estimation algorithms are used to avoid high cost of true depth data, then depth estimation becomes feasible, but the algorithms require complex network training and optimization
Solution Approach 1:
The system uses self-supervised learning where the depth estimation network trains using synthetic data and self-generated supervisory signals. The network learns to estimate depth by minimizing reconstruction error between synthesized and actual images, enabling depth data generation without costly real annotations while managing training complexity through automated self-training mechanisms.
Solution Approach 2:
The patent performs preliminary network training using synthetic data and pre-computed ground truth before deployment. This pre-training phase prepares the depth estimation model with foundational knowledge, reducing the complexity of online optimization during actual robot operation and enabling efficient real-time depth estimation.
Data Source
AI summary
Disclosed is a method for constructing a bionic binocular active visual simultaneous localization and mapping (SLAM) system. The method includes: S1, constructing a binocular active visual system, peripheral vision is simulated using a panoramic camera, a panoramic value map is constructed based on the panoramic camera, and a binocular bionic eye gaze control strategy based on the panoramic value map is proposed; S2, designing a SLAM algorithm framework which is suitable for motion of the two bionic eyes, where relative pose optimization of left and right cameras based on a Perspective-n-Point algorithm and depth calculation for binocular matching points based on triangulation are adopted in a SLAM tracking thread; and S3, with the assistance of camera poses estimated by the SLAM system and sparse landmark point coordinates, training a self-supervised depth estimation network, to provide dense depth estimation for a mapping module of the SLAM system, achieving dense mapping.