Bird's-Eye-View Image Border Width Adjustment for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vehicle-periphery-monitoring systems face challenges in accurately identifying the shape and position of objects due to camera range changes and blind spots, making it difficult for drivers to monitor the vehicle's periphery effectively.
Innovation Solution
An image-processing system that generates bird's-eye-view images with overlapping end portions, combines them into a composite image, and adjusts the border image width based on object detection and vehicle posture changes to enhance object visibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple bird's-eye-view images are combined with mask patterns at joining portions, then object detection capability is improved, but the ability to identify object shape and specific position deteriorates
Solution Approach 1:
The patent extracts the border image from the overlap portion and separates it from the main bird's-eye-view image. By taking out the border region and processing it independently with dynamic width adjustment, the system maintains object detection capability while improving shape and position identification accuracy in the main image areas.
Solution Approach 2:
The patent applies dynamic adjustment to the border image width based on object detection results. When an object is detected near a joining portion, the border width is increased to enhance visibility and provide more information about the object's shape and position, thereby resolving the contradiction between detection reliability and identification precision.
2Area of stationary object
If camera range is adjusted according to vehicle posture changes, then monitoring coverage is improved, but blind spots may occur between adjacent image regions
Solution Approach 1:
The patent performs preliminary action by pre-defining overlap portions between adjacent bird's-eye-view images before combining them. This ensures that even when camera ranges shift due to vehicle posture changes, the predetermined overlap regions maintain continuous coverage and prevent blind spots from forming between image regions.
Solution Approach 2:
The patent addresses the two-dimensional gap problem by utilizing the third dimension of image processing - creating a border image that extends into the depth dimension at joining portions. This border image fills the gap between adjacent images, ensuring continuous monitoring coverage across the entire vehicle periphery.
3Device complexity
If fixed-width border images are used at joining portions, then system complexity is reduced, but adaptability to object detection needs deteriorates
Solution Approach 1:
The patent implements dynamic border width adjustment where the border image width at joining portions is not fixed but changes based on object detection results. When objects are detected near joining portions, the border width is automatically increased to enhance object visibility and provide better spatial context, making the system adaptable to different detection scenarios.
Solution Approach 2:
The system employs feedback mechanisms where object detection results from the main image areas are used to adjust the border image width at joining portions. This feedback loop allows the system to adaptively enhance object visibility by modifying border characteristics based on real-time detection needs, improving versatility without excessive complexity.
Data Source
AI summary
A vehicle-periphery-image-providing system may include an image-capturing unit, a viewpoint-change unit, an image-composition unit, an object-decttion unit, a line-width-setting unit, and a line-selection unit. The image-capturing units, such as cameras, capture images outside a vehicle periphery and generate image-data items. The viewpoint-change unit generates a bird's-eye-view image for each image-data item based on the image-data item so that end portions of the real spaces corresponding to two adjacent bird's-eye-view images overlap each other. The image-composition unit generates a bird's-eye-view-composite image by combining the bird's-eye-view images according to a predetermined layout. The object-detection unit detects an object existing in the real space corresponding to a portion where the bird's-eye-view images of the bird's-eye-composite image are joined to each other. The line-width-setting unit sets the width of the line image corresponding to the joining portion. The line-selection unit adds a line image having the set width to an overlap portion of one of the bird's-eye-view images.