Visual Intersection Recognition Using CNN Place IDs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous navigation systems face challenges in recognizing intersections in GPS-denied outdoor environments, particularly in urban canyons and open outdoor settings, where GPS signals are weak or unavailable, and LIDAR-based solutions are ineffective due to distance limitations.
Innovation Solution
A system utilizing a visual sensor and a computing device with a convolutional neural network (CNN) that processes images from RGB cameras to predict place IDs along a planned path, employing a temporal window to confirm IDs based on sequential image analysis, allowing for accurate navigation even in environments where traditional methods fail.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If GPS-based intersection recognition is used, then the system works well in open environments with strong GPS signals, but it fails in GPS-denied outdoor environments such as urban canyons and tunnels
Solution Approach 1:
The patent introduces visual sensors (cameras) as an intermediary system to capture images of the environment, which are then processed by CNN models to recognize intersections and determine place IDs. This intermediary visual recognition system bridges the gap in GPS-denied environments by providing an alternative method for intersection detection that does not rely on satellite signals.
Solution Approach 2:
The patent replaces the GPS-based mechanical/electronic positioning system with a vision-based recognition system. Instead of relying on GPS satellites and signal reception, the system uses cameras to capture visual data and CNN algorithms to process and interpret the images for intersection detection, effectively substituting one detection mechanism with another that is more adaptable to GPS-denied environments.
2Measurement precision
If LIDAR-based intersection recognition is used, then it shows good performance in closed outdoor environments, but it fails in open outdoor environments due to distance limitation
Solution Approach 1:
The patent replaces the LIDAR optical ranging system with a vision-based system using RGB cameras and CNN processing. This substitution allows the system to overcome LIDAR's distance limitations by using image processing techniques that can recognize intersections and place IDs from greater distances and in more varied lighting conditions, thereby extending the effective detection range.
3Reliability
If visual sensors with CNN are used for intersection recognition, then the system achieves good performance in GPS-denied environments, but it requires processing multiple images with temporal windows which increases computational complexity
Solution Approach 1:
The patent applies preliminary action by pre-training CNN models with large datasets of labeled images before deployment. The models are pre-trained to recognize various intersection types, road signs, and environmental features, which reduces the computational burden during real-time operation. Additionally, the system uses a temporal window approach that processes images in batches rather than individually, optimizing computational efficiency.
Solution Approach 2:
The patent uses a temporal window that processes a predetermined number of following images (e.g., 3-5 images) to confirm place ID predictions. This partial action approach balances accuracy and computational load by not requiring excessive image processing while still achieving reliable intersection recognition through multiple image validation.
Data Source
AI summary
A system and method for autonomous navigation. The system includes a computing device having a processor and a storage device storing computer executable code. The computer executable code, when executed at the processor, is configured to: provide a planned path having intersections in an environment, where the intersections and roads therebetween are represented by sequential place identifications (IDs); receive images of the environment; perform convolutional neural network on the images to obtain predicted place IDs; when a predicted place ID of a current image is next to a place ID of a previous image, and is the same as predicted place IDs of a predetermined number of following images, define the predicted place ID as place IDs of the current and the following images; and perform autonomous navigation based on the planned path and the image place IDs.