Object Recognition Device Adaptive Region Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional wide-angle lenses used for imaging both long-distance and close-proximity regions often result in degraded image quality due to varying incident angles, affecting subsequent image processing.
Innovation Solution
An imaging lens with a characteristic inflection point in the change rate of incident angle per image height, allowing for high angular resolution in central regions for long-distance imaging and lower resolution in peripheral regions for wide-angle views, combined with adaptive parameter adjustments for image recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a conventional wide-angle lens is used to image both long-distance and close-proximity regions, then a wide field of view is achieved, but image quality is degraded in portions of the image due to varying incident angles
Solution Approach 1:
The image region is segmented into a first region (central area with small incident angles) and a second region (peripheral area with large incident angles). Different processing approaches are applied to each region: the first region uses standard processing, while the second region uses processing adapted to handle degraded image quality, thereby resolving the contradiction between wide field of view and image quality
Solution Approach 2:
Different quality standards and processing methods are applied to different parts of the image. The central region maintains high image quality requirements, while the peripheral region uses adjusted processing parameters suitable for its degraded quality characteristics, allowing the system to achieve both wide coverage and acceptable quality throughout
2Measurement precision
If high angular resolution is used in central regions for long-distance imaging, then long-distance recognition accuracy is improved, but peripheral regions with wide-angle views suffer from lower resolution
Solution Approach 1:
The processing is segmented by region: the first region (central) undergoes processing optimized for high angular resolution and long-distance accuracy, while the second region (peripheral) undergoes processing adapted to its lower resolution characteristics, allowing each region to be optimized for its specific requirements
Solution Approach 2:
Processing parameters are changed based on region and distance: for the first region, parameters are set for high precision long-distance recognition, while for the second region, parameters are adjusted to accommodate wide-angle viewing with lower resolution, effectively resolving the precision-resolution trade-off
3Device complexity
If a single processing method is applied to the entire image, then processing simplicity is maintained, but recognition accuracy decreases in regions with degraded image quality
Solution Approach 1:
The processing method is segmented into at least two types: a first processing method applied to the first region and a second processing method applied to the second region. This segmentation allows the system to maintain good recognition accuracy in both regions while keeping the overall processing framework relatively simple through automated region-based method selection
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
The purpose of the present invention is to obtain reliable recognition performance even in a lens which is capable of long-distance and close proximity photography. This object recognition device 1 comprises an image acquisition unit which acquires an image 1000 which includes a plurality of image regions 1001, 1002, 1003 which have differing resolutions, and a recognition processing unit which carries out a recognition process upon an object within the screen 1000. One image region 1003, among the plurality of image regions 1001, 1002, 1003 of the screen 1000, has a lower resolution than the other image regions 1001, 1002. The recognition processing unit carries out the recognition process of the object on the basis of assessment references which differ among the one image region 1003 and the other image regions 1001, 1002.