3D Object Detection from Overhead Difference Images Under Shadow Interference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing three-dimensional object detection systems face accuracy issues when shadows of structures like buildings and signs appear between a vehicle and nearby objects, affecting the detection of three-dimensional objects during travel.
Innovation Solution
A three-dimensional object detecting device that converts overhead images captured at different times into aligned difference images, generates a masked difference image to mask non-object regions, and identifies near and far ground contact lines to accurately determine the location and width of three-dimensional objects, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If overhead images are used for three-dimensional object detection, then detection capability is provided, but detection accuracy deteriorates when shadows are present between the vehicle and objects
Solution Approach 1:
The patent segments the overhead image into multiple regions based on shadow detection. The image processing section divides the detection area into shadow regions (where shadows are detected) and non-shadow regions, applying different processing strategies to each segment. This segmentation allows the system to exclude shadow-affected areas from object detection, thereby maintaining high detection accuracy in the presence of shadows.
Solution Approach 2:
The patent extracts and removes shadow regions from the overhead image before performing three-dimensional object detection. The shadow detection section identifies shadow areas, and these extracted shadow regions are then excluded from the detection process. This extraction approach eliminates the harmful effect of shadows on detection accuracy while preserving the integrity of non-shadow regions for accurate object detection.
2Measurement precision
If difference images are generated from overhead images captured at different times, then three-dimensional object detection is enabled, but detection accuracy deteriorates due to shadow noise in the difference image
Solution Approach 1:
The patent performs preliminary shadow detection and image segmentation before generating the difference image. By detecting shadows in advance and segmenting the overhead images accordingly, the system ensures that shadow regions are excluded from the difference image generation process. This preliminary action prevents shadow noise from being incorporated into the difference image, thereby maintaining high detection accuracy for three-dimensional objects.
Solution Approach 2:
The patent converts the harmful effect of shadows into a beneficial filtering mechanism. By detecting shadow patterns and using them to define exclusion zones, the system transforms shadow interference into a useful tool for improving detection accuracy. The shadow regions, which would normally degrade detection quality, are instead used to guide the segmentation process and enhance the reliability of object detection in non-shadow areas.
3Productivity
If the entire difference image is processed for object detection, then comprehensive detection is achieved, but processing efficiency deteriorates due to unnecessary computation in shadow regions
Solution Approach 1:
The patent segments the overhead image into shadow and non-shadow regions before object detection processing. This segmentation allows the system to restrict subsequent detection algorithms to only the non-shadow regions, eliminating unnecessary computation in shadow areas. As a result, processing efficiency is significantly improved while maintaining comprehensive detection coverage in valid regions.
Solution Approach 2:
The patent applies partial processing by focusing computational resources only on non-shadow regions rather than processing the entire image. This selective approach performs detection only where it is meaningful and effective, avoiding the time waste of analyzing shadow regions where objects cannot be reliably detected. The partial action principle optimizes productivity by concentrating computational effort on productive areas.
Data Source
AI summary
A three-dimensional object detecting device generates a mask image that masks regions outside a three-dimensional object candidate region in a difference image of a first overhead image and a second overhead view image for which the imaging locations O are mutually aligned, identifies a near ground contact line of a three-dimensional object based on a masked difference image. The difference image is masked with the mask image, finds an end point of the three-dimensional object based on the masked difference image, identifies the width of the three-dimensional object based on a distance between a non-masking region boundary and the end point of the three-dimensional object in the mask image, identifies a far ground contact line of the three-dimensional object based on the width of the three-dimensional object and the near ground contact line, and identifies the location of the three-dimensional object in the difference image based on the near ground contact line and the far ground contact line.