Semantic Point Cloud Object Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current vehicle computer systems face challenges in accurately determining the physical characteristics of objects in their environment, such as shape and orientation, due to limitations in bounding box data, which hinders effective path determination and object avoidance in autonomous or semi-autonomous driving.
Innovation Solution
The system generates a transformed semantic point cloud image using a variational autoencoder neural network, combining image and radar data to provide a representation of objects from their perspective, allowing for improved determination of physical characteristics like dimensions, pose, and contour, enabling more accurate 3D bounding box creation and path planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a bounding box is used to represent an object, then the object detection process is simplified, but the physical characteristics such as shape, orientation, and contour cannot be determined
Solution Approach 1:
The patent segments the object representation into multiple components: a bounding box for basic localization and a point cloud for detailed physical characteristics. This segmentation allows the system to maintain the simplicity of bounding box detection while adding point cloud data to capture shape, orientation, and contour information that the bounding box alone cannot provide.
Solution Approach 2:
The patent transitions from a two-dimensional bounding box representation to a three-dimensional point cloud representation. This dimensional enhancement allows the system to capture depth information and physical characteristics such as shape, orientation, and contour, resolving the limitation of the bounding box approach.
2Measurement precision
If more data is collected to determine physical characteristics, then object identification accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent introduces a neural network as an intermediary that automatically processes the point cloud data to extract physical characteristics such as shape, orientation, and contour. This intermediary handles the complex data processing tasks, reducing the computational burden on the overall system while maintaining high object identification accuracy.
Solution Approach 2:
The patent creates a simplified representation (point cloud) that copies the essential physical characteristics of the object without requiring full complexity of the original data. This point cloud copy contains the necessary information for determining shape, orientation, and contour while being more manageable for processing than complete sensor data.
Data Source
AI summary
Upon identifying image data associated with an object, radar data associated with the object is identified. A semantic point cloud image is generated based on the image data and the radar data. Transformed semantic point cloud image from a perspective of the object is determined with a variational auto-encoder neural network trained to accept the semantic point cloud image of the object and to generate the transformed semantic point cloud image from the perspective of the object. Physical characteristics of the object are determined based on the transformed semantic point cloud image.