Object Detection Using Rotating Haar Feature Database
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection algorithms, such as the Viola-Jones algorithm, require multiple Haar-feature databases for different orientations, leading to increased computational time and complexity when detecting objects in images with varying orientations.
Innovation Solution
The method involves capturing images in sampling windows and using a classifier with rotating sub-classifiers to detect objects in multiple orientations using a single Haar-Feature Database (HFDB), reducing the need for multiple orientations and halving the detection time by rotating the classifier and image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple Haar-feature databases are used for different orientations, then object detection accuracy for multiple orientations is improved, but computational time and algorithm complexity increase
Solution Approach 1:
The patent makes a single Haar-feature database serve multiple orientations by rotating the database at predetermined angles (0°, 45°, 90°, 135°) to detect objects in different orientations. This eliminates the need for separate Haar-feature databases for each orientation, reducing computational overhead while maintaining detection accuracy across multiple orientations.
Solution Approach 2:
The patent dynamically rotates the Haar-feature database by predetermined angles during the detection process to adapt to objects in different orientations. Instead of using static, orientation-specific databases, the system dynamically adjusts the database orientation to match the object's orientation, improving efficiency while maintaining accuracy.
2Measurement precision
If multiple Haar-feature databases are used for different orientations, then object detection accuracy for multiple orientations is improved, but device complexity increases
Solution Approach 1:
The patent makes a single Haar-feature database serve multiple orientations by rotating the database at predetermined angles (0°, 45°, 90°, 135°) to detect objects in different orientations. This eliminates the need for separate Haar-feature databases for each orientation, reducing computational overhead while maintaining detection accuracy across multiple orientations.
3Loss of time
If a single Haar-feature database is used with rotation, then detection time is reduced, but the ability to detect objects in all orientations may be compromised
Solution Approach 1:
The patent dynamically rotates the Haar-feature database by predetermined angles during the detection process to adapt to objects in different orientations. Instead of using static, orientation-specific databases, the system dynamically adjusts the database orientation to match the object's orientation, improving efficiency while maintaining accuracy.
Solution Approach 2:
The patent applies periodic rotation of the Haar-feature database at predetermined angles (0°, 45°, 90°, 135°) to cover all major object orientations. This periodic angular sampling ensures comprehensive orientation coverage while maintaining efficient detection speed, balancing accuracy and performance.
Data Source
AI summary
A method for object detection and an apparatus using the same are provided, and the method includes: An image is captured, in which the image includes a plurality of sampling-windows. A first-stage sub-classifier of a classifier is used to detect whether the sampling-windows contain an object therein. The classifier is rotated at least one time by a predetermined rotation angle and the first-stage sub-classifier of the classifier is used to detect whether the sampling-windows contain the object after each rotating, wherein when the object is detected within the sampling-windows, keep detecting whether the sampling-windows contain the object therein sequentially by a second-stage sub-classifier to an Nth stage sub-classifier of the classifier with the same orientation. The image is rotated at least one time by a predetermined image angle and the above-mentioned operations of detecting the object is performed after each rotating. The sampling-windows containing the object are output.