Object Detection Using Dynamic Brightness Gradient Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems face challenges in maintaining robustness against environmental changes and high processing costs when recognizing multiple objects using local feature values and affine transformations, leading to decreased recognition performance due to information degradation and noise.
Innovation Solution
An object detection system that calculates feature values from brightness gradients in peripheral areas, dynamically setting sampling points and areas based on object conditions, allowing for robust object recognition with reduced processing costs and quick response times.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If local feature values and affine transformations are used for recognizing multiple objects, then recognition capability is improved, but processing cost increases and recognition performance decreases due to information degradation and noise
Solution Approach 1:
The patent extracts only the essential brightness gradient information from image areas rather than performing comprehensive affine transformations on entire images. By extracting feature values solely from brightness gradients in peripheral areas around sampling points, the system avoids the computational overhead of full affine transformations while retaining key object characteristics for recognition.
Solution Approach 2:
The patent divides the image into multiple discrete areas around sampling points and calculates brightness gradients independently for each area. This segmentation approach allows parallel processing of multiple regions, reducing overall processing cost while maintaining the ability to recognize multiple objects simultaneously through aggregated feature values.
2Adaptability or versatility
If affine transformation is applied to images, then object recognition under various conditions is improved, but information degradation and noise increase leading to decreased recognition performance
Solution Approach 1:
The patent changes the approach from transforming image coordinates through affine transformations to directly calculating brightness gradients in peripheral areas. This parameter change from spatial transformation to gradient computation preserves information by working directly with intensity variations that are inherently robust to environmental changes like lighting and viewpoint, avoiding the information loss associated with coordinate transformations.
3Productivity
If a small number of sampling points are used, then processing speed is improved, but recognition accuracy may decrease
Solution Approach 1:
The patent merges information from multiple sources to compensate for using fewer sampling points: (1) combining brightness gradient data from multiple peripheral areas around each sampling point, (2) aggregating feature values across all sampling points, and (3) integrating results from multiple object candidates. This merging strategy maintains recognition accuracy despite reduced sampling point count by accumulating evidence from diverse sources.
Data Source
AI summary
According to an embodiment, an object detection system includes an obtaining unit, an estimating unit, a setting unit, a calculating unit, and a detecting unit. The obtaining unit is configured to obtain an image in which an object is captured. The estimating unit is configured to estimate a condition of the object. The setting unit is configured to set, in the image, a plurality of areas that have at least one of a relative positional relationship altered according to the condition and a shape altered according to the condition. The calculating unit is configured to calculate a feature value of an image covering the areas. The detecting unit is configured to compare the calculated feature value with a feature value of a predetermined registered object, and detect the registered object corresponding to the object.