RGB-D Dynamic Obstacle Detection for MAV Navigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing obstacle detection systems using RGB-D sensors struggle to effectively detect and track dynamic obstacles in cluttered environments, particularly for robot navigation, as they often require high computational resources and are not suitable for Micro Aerial Vehicles (MAVs).
Innovation Solution
A system and method that utilize RGB-D sensors to capture depth images, construct u-depth maps, and detect obstacles by converting these maps into binary images, estimating obstacle width and depth range, and tracking obstacles by matching signature closeness in subsequent images, while estimating dynamic obstacles based on velocity thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional obstacle detection systems use RGB-D sensors with high computational processing, then obstacle detection accuracy is improved, but computational cost and processing time increase making it unsuitable for MAVs
Solution Approach 1:
The patent segments the obstacle detection process into distinct stages: depth image acquisition, binary image conversion using dynamic thresholding, obstacle candidate identification, and verification. This segmentation allows each stage to be optimized independently, reducing overall computational complexity while maintaining detection accuracy
Solution Approach 2:
The patent extracts only the essential features needed for obstacle detection from depth images by converting them to binary images through dynamic thresholding. This extraction process removes unnecessary data while preserving critical obstacle information, significantly reducing computational requirements
2Adaptability or versatility
If dynamic obstacle detection is implemented in cluttered environments, then detection capability is improved, but false detections and processing complexity increase
Solution Approach 1:
The patent performs preliminary conversion of depth images to binary images using dynamic thresholding before obstacle identification. This preliminary action simplifies the image data structure and reduces noise, making subsequent obstacle detection more accurate and less prone to false positives in cluttered environments
Solution Approach 2:
The patent implements a verification mechanism where detected obstacles are validated against expected characteristics. This feedback loop filters out false detections by comparing detected objects against known obstacle properties, improving reliability in cluttered environments
3Speed
If real-time obstacle tracking is implemented, then navigation responsiveness is improved, but computational resources are consumed
Solution Approach 1:
The patent applies partial action by tracking only verified obstacles that meet detection criteria, rather than processing all detected objects. This selective approach reduces computational energy consumption while maintaining real-time tracking capability for relevant obstacles
Solution Approach 2:
The patent discards incomplete or unverifiable obstacle data that does not meet detection thresholds, focusing computational resources only on valid obstacles. This filtering approach reduces energy consumption while maintaining tracking effectiveness
Data Source
AI summary
This disclosure relates generally to method and system to detect and estimate dynamic obstacles using Red Green Blue-Depth RGB-D sensors for robot navigation. Obstacle detection and tracking tasks with efficient processing of depth images becomes complex in dynamic and cluttered environments. The method detects one or more obstacles being identified in the one or more depth images in two dimensional space using the depth map captured using a Red Green Blue-Depth RGB-D sensor configured to a mobile robot. Further, a restricted v-depth map is computed between specified columns taken from the width of each dynamic obstacle of the u-depth map and within the depth range derived from the u-depth map. The one or more dynamic obstacles are tracked by matching segments of subsequent depth images to predict future position in the region of the monitoring environment by computing a closeness signature. The method evaluates correctness with open data sequence and real world data.