Monocular Object Pose Estimation Using Anchor Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection technologies using monocular cameras struggle to accurately estimate the pose of objects, particularly in scenarios where initial assumptions do not apply, leading to reduced reliability in vehicle control systems.
Innovation Solution
An object detecting apparatus comprising a detector and an estimation processing unit that uses a neural network to estimate the pose of objects by identifying anchor points and orientation types within a captured image, enabling accurate detection of obstacles and other vehicles, pedestrians, and roadside objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional object detection technology using monocular cameras is used, then the device complexity is low, but the measurement precision of object pose is insufficient
Solution Approach 1:
The patent segments the object detection task into multiple independent components: detecting object candidate areas, estimating anchor point positions, and determining orientation types. This segmentation allows each component to be processed separately with dedicated algorithms, improving overall measurement precision without requiring a complete system redesign
Solution Approach 2:
The patent introduces anchor points as an intermediate dimensional representation between the 2D image plane and the 3D object pose. By estimating anchor point positions and using them to infer orientation types, the system bridges the gap between 2D image data and 3D pose estimation, significantly improving measurement precision
2Reliability
If assumptions are made for object pose estimation, then the device complexity is reduced, but the reliability of vehicle control is compromised
Solution Approach 1:
The patent changes the parameters used for pose estimation by introducing anchor point positions and orientation types as intermediate parameters. Instead of directly estimating pose from image data under assumptions, the system estimates anchor points first, then uses these to determine orientation types, providing more reliable results for vehicle control
Solution Approach 2:
The patent introduces anchor points as intermediary elements that mediate between the image data and the final pose estimation. These anchor points serve as reliable reference markers that connect the 2D image plane to the 3D object pose, improving reliability without requiring complex assumption-based models
3Measurement precision
If multiple parameters are estimated for object pose, then the measurement precision is improved, but the loss of information increases
Solution Approach 1:
The patent performs preliminary estimation of anchor point positions before determining orientation types. This preliminary action provides accurate reference information that reduces information loss in the subsequent orientation estimation step, as the anchor points serve as reliable anchors for the final pose calculation
Data Source
AI summary
According to an embodiment, an object detecting apparatus includes a detector and an estimation processing unit. The detector is configured to detect an object candidate area from a captured image. The estimating processing unit is configured to, by performing estimation processing using a part of or a whole of the captured image including at least the object candidate area, output object information including at least information representing a pose of an object in the object candidate area. The estimation processing includes: a first process estimating, from among vertices of a cuboid circumscribing the object and making contact with a road surface, positions of at least two vertices on the captured image that are viewable from a viewpoint of the captured image; and a second process estimating to which one of right-front, left-front, right-rear, or left-rear of the object the vertices having positions estimated by the first process are respectively corresponding.