3D Object Detection Using Neural Network Template Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current navigation systems face challenges in accurately detecting three-dimensional objects using remote sensing methodologies, which suffer from accuracy constraints and limitations in determining object orientation and location.
Innovation Solution
A method and system utilizing a deep neural network to process sensor information, such as LIDAR point clouds, to generate a three-dimensional representation of the environment, identify object features, and compare them to pre-defined templates to determine object locations and rotational orientations, using rotated bounding boxes with length, width, height, and rotation values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If remote sensing methodologies are used to detect objects, then navigation applications can be supported, but detection accuracy is constrained
Solution Approach 1:
The patent transitions from two-dimensional sensor data to three-dimensional object representation by introducing height dimension and rotational orientation parameters. Objects are represented as 3D bounding boxes with dimensions (x, y, z) and orientation (roll, pitch, yaw), enabling accurate detection of three-dimensional objects in navigation environments.
Solution Approach 2:
The system changes the parameter representation from simple 2D coordinates to comprehensive 3D parameters including position (x, y, z), size (length, width, height), and orientation (rotation values). This parameter expansion allows the neural network to capture full spatial characteristics of objects, resolving the accuracy constraint of traditional remote sensing.
2Measurement precision
If traditional object detection methods are used, then processing is simpler, but determination of object orientation and location is limited
Solution Approach 1:
The patent introduces 3D bounding boxes as an intermediary representation between raw sensor data and object characteristics. These bounding boxes serve as a mediator that encapsulates position, size, and orientation information, making it easier for the neural network to determine object location and orientation without directly complex calculations.
Solution Approach 2:
The system performs preliminary action by pre-defining 3D bounding box templates with various orientations and dimensions before detection. These templates are prepared in advance and used to match against detected objects, enabling efficient determination of object orientation and location without real-time complex computations.
3Reliability
If basic sensor data processing is used, then computational resources are saved, but object detection accuracy is reduced
Solution Approach 1:
The patent segments the object detection process into distinct stages: sensor data acquisition, 3D space generation, feature identification by neural network, template matching, and parameter determination. This segmentation allows each component to be optimized independently, maintaining high detection reliability while managing computational resources efficiently through specialized processing at each stage.
Data Source
AI summary
A method for detecting objects in an environment includes obtaining, from one or more sensors, distance measurements from the one or more sensors to portions of the environment, generating a representation of a three-dimensional space using the distance measurements, identifying object features in the representation of the three-dimensional space using a neural network, comparing the object features to pre-defined three-dimensional templates to generate scores that represent correspondence of the object features to the pre-defined three-dimensional templates, and determining a location and a rotational orientation for a three-dimensional object based on the scores.