Occupancy Grid CNN Parsing for Real-Time Driving Context
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining a driving context in outdoor autonomous driving environments fail to effectively handle complex and dynamic scenes, particularly due to the interaction with dynamic objects and lack of consideration for sensor uncertainties.
Innovation Solution
A method utilizing occupancy grids and convolutional neural networks to process sensor data from Sonar, Lidar, and Radar, converting the grids into image representations for real-time classification, employing the Dempster-Shafer theory for sensor fusion and gradually updating grid content, with a simplified neural network architecture for efficient driving context determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If occupancy grids are constructed from sensor data in outdoor autonomous driving environments, then the system can represent the environment structure, but the complexity of dynamic scenes and interaction with dynamic objects makes accurate classification difficult
Solution Approach 1:
The system segments the complex outdoor environment into discrete occupancy grid cells, where each cell represents a simplified unit of space. This segmentation transforms the continuous complex scene into a structured grid that can be processed systematically, reducing the overall system complexity while maintaining classification accuracy.
Solution Approach 2:
The patent transforms the 3D spatial environment into a 2D occupancy grid representation, adding an abstract dimensional layer. This dimensionality change simplifies the processing of complex 3D scenes by projecting them onto a 2D grid plane, making the system more manageable while preserving essential spatial relationships for accurate classification.
2Productivity
If traditional neural networks are used for parsing occupancy grids, then the system can process environment data, but the computational requirements are too high for real-time application on low-performance processors
Solution Approach 1:
The patent extracts and implements only the essential computational components needed for occupancy grid parsing, removing unnecessary complex layers from traditional neural networks. This extraction creates a streamlined network architecture that maintains real-time processing capability while significantly reducing computational energy consumption, making it suitable for low-performance processors in autonomous vehicles.
3Adaptability or versatility
If occupancy grid content is updated frequently to track dynamic objects, then the system can adapt to changing environments, but the processing time and computational load increase
Solution Approach 1:
The system implements periodic updates of occupancy grid content at optimized intervals rather than continuous updates. This periodic action allows the system to adapt to changing environments and track dynamic objects effectively while maintaining regular processing cycles that prevent excessive computational load and minimize processing time loss.
Data Source
Figure 1~2
Figure 3~4
Figure 5(a)~5(c)
AI summary
The present invention is related to a method, a computer program, and an apparatus for determining a driving context of a vehicle. In a first step, sensor data of one or more sensors of the vehicle are received (10). Then an occupancy grid is determined (11) based on the sensor data. Finally, the occupancy grid is parsed (12) with a convolutional neural network for determining the driving context.