Shared SGM Environment Modeling for Extended Vehicle Perception
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle/robot systems are limited to constructing environment models based solely on their own sensor data, restricting the range of the model to their immediate surroundings.
Innovation Solution
A method and device that generate a Signatured Gaussian Mixture Model (SGM) based on local sensor data and combine it with SGM models from other vehicles/robots to create an enlarged environment model, utilizing shared sensor data via internet or direct connections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional map data or LIDAR data is used to represent the environment, then the environment model can be constructed, but the system requires expensive specialized equipment and complex processing hardware
Solution Approach 1:
The patent uses camera images (optical copies) instead of direct LIDAR point cloud data to represent the environment. The neural network processes these visual copies to generate occupancy grids and environment models, replacing expensive LIDAR hardware with standard camera equipment while maintaining environmental representation capability.
Solution Approach 2:
The patent replaces mechanical LIDAR scanning systems with an optical-based vision system using standard cameras and neural network processing. This substitution eliminates the need for complex mechanical scanning hardware while achieving environment modeling through software-based image processing.
2Measurement precision
If LIDAR data is processed to create an environment model, then accurate spatial information is obtained, but the processing requires significant computational resources and time
Solution Approach 1:
The system pre-processes camera images by detecting features and generating initial occupancy grid representations before full environment modeling is needed. This preliminary processing reduces the computational burden during real-time operation and accelerates the overall modeling process.
Solution Approach 2:
The patent processes only the necessary portions of image data at each processing stage, using selective feature detection and region-based processing rather than analyzing entire high-resolution images. This partial processing approach maintains spatial accuracy while reducing computational time and resource requirements.
3Adaptability or versatility
If multiple data sources (LIDAR, maps, cameras) are integrated to build a comprehensive environment model, then the model completeness is improved, but the system complexity and cost increase
Solution Approach 1:
The patent makes the vision system multi-functional by using standard cameras to perform multiple tasks: capturing images for environment modeling, detecting features for localization, and providing input for navigation decisions. This eliminates the need for separate specialized sensors for each function, reducing overall system complexity while maintaining completeness.
Solution Approach 2:
The patent introduces an intermediary neural network processing layer that integrates information from multiple sources (camera images, map data, localization data) and transforms them into a unified environment representation. This intermediary processing layer simplifies the integration of multiple data sources by providing a standardized interface and coordinated processing approach.
Data Source
Figure 1
Figure 2~4
Figure 5~6
AI summary
The present invention provides a computer-implemented method and a data processing device for constructing an environment model. The method comprises: generating a first Signatured Gaussian Mixture, SGM, model corresponding to a first part of the environment based on a first sensor data; receiving a second SGM model corresponding to a second part of the environment; and constructing a third SGM model comprising the first SGM model and the second SGM model.