Vehicle Object Estimation for Oblique Posture Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object estimation systems for vehicles, such as those disclosed in Japanese Unexamined Patent Application Publication (JP-A) No. 2019-008460, face significant errors in estimating the location of objects, particularly when the object is in an oblique posture, due to the assumption of a cubic shape outline, leading to potential collisions as vehicles with rounded outlines are not accurately represented.
Innovation Solution
The system employs a deep learning neural network that learns the relevance between captured images and estimation frames, estimating an image region widened from the object's image region when it is in an oblique posture, to improve location estimation accuracy by using a two-dimensional frame or circumscribed cube that encloses the object, providing more accurate position, direction, and distance data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a cubic shape outline assumption is used for object estimation, then the estimation process is simple, but the estimation accuracy deteriorates when objects have rounded outlines or are in oblique postures
Solution Approach 1:
The patent transitions from two-dimensional image region estimation to three-dimensional space region estimation by introducing depth information and spatial coordinates. The estimation frame is extended from a 2D bounding box to a 3D volume that accounts for the object's actual spatial occupation, thereby resolving the inaccuracy caused by assuming cubic shapes in 2D images.
Solution Approach 2:
The patent changes the estimation parameters from simple 2D coordinates (x, y) to 3D spatial parameters (x, y, z) including depth information. It also introduces parameters for object posture and orientation, allowing the estimation system to adapt to objects in various orientations and postures, thereby improving accuracy without significantly increasing computational complexity.
2Measurement precision
If the image region is not widened for oblique posture objects, then the estimation is more precise for the visible object boundary, but the vehicle may come into contact with the object due to underestimated spatial occupation
Solution Approach 1:
The patent applies preliminary anti-action by pre-expanding the estimation frame to include the full spatial occupation volume of objects in oblique postures. This preventive expansion ensures that the vehicle's path planning system has advance warning of the object's complete spatial extent, preventing potential collisions before they occur.
Solution Approach 2:
The patent introduces a safety margin or cushioning zone around the estimated object volume, particularly for objects in oblique postures. This beforehand cushioning creates a buffer region that accounts for uncertainties in estimation and the object's actual spatial occupation, ensuring reliable collision avoidance even when estimation has some error.
Data Source
AI summary
An object estimation device includes an acquisition unit and an estimation unit. The acquisition unit acquires a space image including an object present in a space. The estimation unit estimates image region data on the basis of a portion or all of the space image including the object. The image region data indicates a location of the object in the space image. On the condition that the object included in the space image acquired by the acquisition unit is included in an oblique posture, the estimation unit estimates an image region widened from an image region of the object included in the oblique posture in the space image, as the image region data.