Overhead View Image Processing for Obstacle Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing technologies for overhead view images of a vehicle's surroundings struggle to accurately depict complex-shaped obstacles like pedestrians or bicycles, making it difficult for drivers to recognize their presence and position, especially when these objects are behind distant obstacles.
Innovation Solution
An image processing device that receives images from multiple directions, associates background objects with common-range images, and generates a display that includes a detection rectangle image with the target object and background objects, allowing drivers to easily recognize the positional relationship of obstacles even if they are not directly visible from their line of sight.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If overhead view images are generated by deforming various images to match the overhead view shape, then the coverage area and field of view are improved, but the recognition accuracy of complex-shaped obstacles deteriorates due to deformation
Solution Approach 1:
The patent segments the overhead view image into multiple regions, each corresponding to a different camera's field of view. Instead of uniformly deforming the entire image, it processes each region separately according to its source camera's characteristics, preserving the original shape information of obstacles in each segment while maintaining comprehensive coverage.
Solution Approach 2:
The patent applies different processing qualities to different regions of the overhead view image. Regions with complex-shaped obstacles maintain higher fidelity to original shapes, while other areas use standard deformation. This local differentiation ensures obstacle recognition accuracy is preserved where needed while maintaining overall coverage.
2Adaptability or versatility
If conventional image deformation methods are used to create overhead view images, then the field of view is expanded, but the ease of recognizing obstacle positions deteriorates due to shape distortion
Solution Approach 1:
The patent adds dimensional information by including both the overhead view image and the original camera images in the display. This multi-dimensional presentation allows drivers to view obstacles from both the synthesized overhead perspective and the original un distorted camera perspectives, enhancing position recognition accuracy.
Solution Approach 2:
The patent introduces an intermediary processing layer that generates region-specific deformation parameters. This intermediary layer selectively applies deformation only where necessary while preserving original shapes in critical regions, mediating between the need for expanded field of view and accurate obstacle recognition.
3Device complexity
If uniform deformation is applied to all images for overhead view generation, then the processing simplicity is maintained, but the manufacturing precision of obstacle shape representation deteriorates
Solution Approach 1:
The patent implements dynamic processing complexity by adapting the deformation approach based on the content of each image region. Regions containing complex-shaped obstacles use minimal or no deformation, while other regions use standard deformation. This dynamic adjustment optimizes the balance between processing complexity and shape accuracy.
Solution Approach 2:
The patent changes the deformation parameters locally across different regions of the overhead view image. Instead of using a single uniform deformation matrix, it adjusts deformation parameters based on the characteristics of each region and its source camera, thereby preserving obstacle shapes in critical areas while maintaining processing efficiency.
Data Source
AI summary
An example image processing device obtains first images capturing a street, on which a target vehicle is running, from a plurality of directions, and position information indicating positions at which the first images are taken, and associates a background object in a second image with a background object in a third image. The second image is an image which the target object is detected from and the third image is an image which is taken by an imaging device having a common imaging range with the second image among imaging devices mounted on the target vehicle. An output image is generated to include the position information indicating a position at which the second image is taken, the image of the detection rectangle which is clipped from the second image and includes the target object and the background object, and the first image of surroundings of the target vehicle.