3D Object Detection Correction via Camera Pose Transformation

Solution Overview

Problem

Conventional 3D object detection systems struggle to perform effectively when adapted to different camera environments, leading to increased data collection costs and reduced accuracy due to the need for labeled learning data across various vehicle types with unique camera settings.

Innovation Solution

A 3D object detection correction apparatus and method that utilizes deep learning to correct unsupervised detection results by accounting for camera position and angle differences between vehicles, employing a processor to generate features from image data and camera information, and adjust detected 3D object positions and angles using pose information of 6 degrees of freedom.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If 3D object detection systems are trained with labeled learning data for each vehicle type, then detection accuracy is improved, but data collection cost and system complexity increase

Engineering Contradiction:

Improvedetection accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and corrects only the camera-related error components (position and angle differences) from the overall detection system, rather than retraining the entire detection model. By isolating and correcting the camera parameter differences through coordinate transformation, the system achieves adaptability across vehicle types without requiring full system reconfiguration or extensive labeled data collection for each vehicle.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the camera parameter values (position and angle) through coordinate transformation based on the difference between target vehicle camera parameters and reference vehicle camera parameters. This parameter adjustment allows the detection model trained on one vehicle type to be adapted to another vehicle type without retraining, thereby reducing data collection costs and system complexity while maintaining detection accuracy.

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If 3D object detection systems are trained with labeled learning data for each vehicle type, then detection accuracy is improved, but data collection cost increases

Engineering Contradiction:

Improvedetection accuracyVSAvoiddata collection cost

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent creates a virtual copy of the reference vehicle's camera parameters and transforms them to match the target vehicle's camera configuration. By copying the detection model and applying coordinate transformation to adapt it to different camera parameters, the system avoids the need to collect and label new data for each vehicle type, significantly reducing data collection costs while maintaining detection accuracy.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent changes the camera parameter values (position and angle) through coordinate transformation based on the difference between target vehicle camera parameters and reference vehicle camera parameters. This parameter adjustment allows the detection model trained on one vehicle type to be adapted to another vehicle type without retraining, thereby reducing data collection costs and system complexity while maintaining detection accuracy.

Inventive Principle:

Principle #35Parameter changes

3Device complexity

If conventional 3D object detection systems are used without correction, then system simplicity is maintained, but detection performance deteriorates in different camera environments

Engineering Contradiction:

Improvesystem simplicityVSAvoiddetection performance

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The patent performs preliminary coordinate transformation to correct camera parameter differences before the detection model processes the input data. By pre-adjusting the camera position and angle parameters based on the known difference between reference and target vehicle configurations, the system ensures accurate detection performance from the start without requiring complex post-processing or adaptive retraining mechanisms.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent changes the camera parameter values (position and angle) through coordinate transformation based on the difference between target vehicle camera parameters and reference vehicle camera parameters. This parameter adjustment allows the detection model trained on one vehicle type to be adapted to another vehicle type without retraining, thereby reducing data collection costs and system complexity while maintaining detection accuracy.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20240203137A1Apparatus and method for correcting detection of 3D objects

Publication Date: 2024.06.20 HYUNDAI MOTOR CO LTD

AI summary

The present disclosure provides a 3D object detection correction apparatus and method. The 3D object detection correction apparatus includes a processor configured to detect a 3D object based on image data. The processor is also configured to correct an error of the detected 3D object based on a difference value of camera information of a vehicle. The 3D object detection correction apparatus also includes a storage configured to store data and algorithms executable by the processor.