Environment Recognition Device Using Color Histograms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional environment recognition devices face challenges in accurately recognizing objects in color images due to overlapping colors, high processing loads, and the inability to synergistically improve recognition accuracy when multiple methods are combined.
Innovation Solution
The device employs an object specifying unit, image data acquiring unit, average color calculating unit, color information adding unit, and color histogram creating unit to recognize objects by calculating average colors of pixel blocks, determining color indexes, and creating one-dimensional color histograms, reducing processing load and enhancing accuracy without area splitting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If area splitting processing is performed on a pixel basis to identify objects, then object recognition capability is improved, but processing load on computer becomes enormous
Solution Approach 1:
The patent divides the image into pixel blocks (groups of pixels) rather than processing individual pixels. Each pixel block is represented by its average color, reducing the data volume while maintaining sufficient information for object recognition. This segmentation approach balances recognition accuracy with processing efficiency.
Solution Approach 2:
The patent applies color histogram analysis selectively to pixel blocks with specific color characteristics rather than all pixels. By focusing processing on relevant color regions and using color indexes to filter data, the system achieves effective object recognition with reduced computational effort.
2Device complexity
If conventional color splitting methods are used to handle overlapping colored objects, then color classification is simplified, but correct area splitting cannot be performed
Solution Approach 1:
The patent transitions from two-dimensional spatial area splitting to one-dimensional color histogram analysis. By accumulating pixel block counts along one direction (e.g., vertical) to create color histograms, the system can identify objects based on color distribution patterns without requiring precise spatial segmentation, thereby solving the overlapping objects problem.
Solution Approach 2:
The patent changes the representation parameter from spatial coordinates to color histogram values. Instead of determining object boundaries in space, the system uses color histogram characteristics (shape, peaks, distribution) to identify and distinguish overlapping objects, achieving accurate recognition without precise area splitting.
3Measurement precision
If multiple recognition methods are used to increase recognition accuracy, then recognition coverage is improved, but synergistic effect to increase accuracy is not achieved
Solution Approach 1:
The patent combines color histogram analysis with pattern matching and knowledge processing into a unified recognition framework. The color histogram provides robust color-based object identification that works synergistically with spatial pattern matching, allowing multiple recognition approaches to complement each other rather than conflict, thereby achieving improved accuracy without excessive complexity.
Data Source
AI summary
A color information creating part calculates the average color of captured image data in units of macro blocks, and determines to which color index the average color of each macro block corresponds. Color arrangement information formed by associating color information with the macro blocks is stored in a color index memory. A histogram creating part creates color histograms representative of the distribution of macro blocks added with color information corresponding to aimed color indexes. Based on the color arrangement information and the color histograms, a color application execution part recognizes objects, surroundings, or the situation of a driver's own vehicle without performing area splitting (shape acquisition) of an image.