Non-perspective Object Detection via Tile Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition techniques fail to accurately identify objects in non-perspective images captured by cameras with fisheye or omnidirectional lenses due to significant distortions, especially for objects farther from the image center.
Innovation Solution
A system that uses a non-perspective object detector to generate an arrangement of tiles based on the camera's field of view, allowing for effective detection and orientation of objects within non-perspective images, including the selection of frequency sub-divisions to enhance detection coverage and display objects upright on a user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If fisheye or omnidirectional lenses are used to capture non-perspective images, then a wide field of view is achieved, but significant distortions occur especially for objects farther from the image center
Solution Approach 1:
The patent divides the non-perspective image into multiple perspective sub-images using a grid pattern overlay. Each cell of the grid corresponds to a perspective view, allowing object detection algorithms to process each segment separately with reduced distortion effects. This segmentation transforms a single distorted wide-angle view into multiple manageable perspective views.
Solution Approach 2:
The patent introduces an intermediary processing step that converts the non-perspective image into multiple perspective sub-images before object detection. This intermediary transformation layer acts as a mediator between the distorted input image and the object detection algorithm, enabling accurate recognition by presenting data in a format suitable for standard detection techniques.
2Ease of operation
If current object recognition techniques are applied directly to non-perspective images, then processing simplicity is maintained, but detection accuracy deteriorates due to image distortions
Solution Approach 1:
The patent segments the complex task of detecting objects in distorted non-perspective images into simpler sub-tasks of detecting objects in multiple undistorted perspective sub-images. This segmentation maintains ease of operation by using standard perspective image processing techniques while improving accuracy through the cumulative results from multiple segments.
Solution Approach 2:
The patent adds a dimensional transformation by converting a single non-perspective image into multiple perspective sub-images arranged in a grid. This dimensional change allows the system to leverage well-established perspective image recognition techniques while handling the complexity of wide-field-of-view imaging.
3Device complexity
If the entire field of view is processed as a single image, then processing steps are minimized, but object detection accuracy decreases due to widespread distortions
Solution Approach 1:
The patent divides the field of view into multiple perspective sub-images through grid-based segmentation. Although this increases the number of processing steps, each segment processes smaller, less distorted regions, improving overall detection accuracy. The segmented approach allows parallel processing of multiple sub-images, mitigating the complexity increase.
Solution Approach 2:
The patent applies partial processing by focusing object detection on specific regions of interest within the segmented grid, rather than uniformly processing the entire field of view. This partial action approach reduces unnecessary processing steps while maintaining detection accuracy for relevant objects.
Data Source
AI summary
Method of detecting objects in non-perspective images starts by generating an arrangement of tiles based on a field of view of a non-perspective camera lens, a predetermined size of the tiles, and a predetermined maximum object radius.The arrangement of the tiles includes the minimum number of tiles to cover the field of view. A non-perspective image is then captured using the non-perspective camera lens. The non-perspective image may be a still image frame or a video. Using the tiles, a plurality of images are generated, respectively, and at least a portion of a first object is detected in one or more images. The first object is generated using the one or more images that include the at least the portion of the first object, and the first object is displayed on a display interface. Other embodiments are described herein.