3D Object Detection Bird's Eye View Edge Stretching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Three-dimensional object detection devices face accuracy issues due to stretching of edges in bird's eye views, leading to low resolution at high positions and misidentification of edges at low positions, resulting in deteriorated detection accuracy.
Innovation Solution
The device calculates luminance differences along vertical imaginary lines in the actual space and detects three-dimensional objects based on continuous luminance differences, unaffected by edge stretching caused by viewpoint conversion, using a luminance difference calculation method that compares pixels near these lines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a differential filter of three pixels by three pixels is used to detect edges, then the detection process is simple, but edges at high positions of three-dimensional objects are missed due to low resolution and edges at low positions are misidentified as noise
Solution Approach 1:
The patent divides the edge detection process into two separate detection stages: first detecting candidate edges using a 3x3 differential filter, then verifying these candidates by checking continuity along radial lines extending from the viewpoint. This segmentation allows the simple filter to operate while adding a verification layer that eliminates false positives and recovers missed edges, thereby resolving the contradiction between simplicity and accuracy.
Solution Approach 2:
The patent introduces an intermediary verification mechanism that checks whether detected edges form continuous patterns along radial lines from the viewpoint. This intermediary step acts as a mediator between the simple differential filter and the final detection result, filtering out noise while preserving and recovering true edges, thus improving accuracy without significantly increasing overall system complexity.
2Area of stationary object
If viewpoint conversion to bird's eye view is performed, then the overall detection coverage is improved, but edges are stretched and resolution varies by height causing detection accuracy to deteriorate
Solution Approach 1:
The patent applies local quality adjustment by adapting the edge detection criteria to different radial distances from the viewpoint. Edges at different heights (different radial distances) are evaluated with appropriate continuity requirements, allowing the system to maintain high detection accuracy across the entire bird's eye view area despite varying edge characteristics at different positions.
Solution Approach 2:
The patent transforms the edge detection problem from a two-dimensional local pixel analysis to a three-dimensional radial continuity check. By evaluating edge continuity along radial lines extending from the viewpoint in 3D space and projecting them to the 2D bird's eye view, the system compensates for the distortion caused by viewpoint conversion and maintains accurate edge detection across all heights.
Data Source
Figure 1
Figure 2(a)~2(b)
Figure 3
AI summary
A three-dimensional object detection device 1 includes a camera 10 and a calculator 20. The calculator 20 performs viewpoint conversion processing on an image captured by the camera 10 to create a bird's eye view image, calculates, for each of a plurality of positions along a vertical imaginary line extending in a vertical direction in an actual space, a luminance difference between two pixels near the position, and detects a three-dimensional object on the basis of continuities of the calculated luminance differences of the respective positions.