Visual Object Detection Saliency Map
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video processing methods, such as background subtraction, often fail to effectively detect objects in highly textured environments due to their local nature, leading to issues like ignoring textured objects and difficulty in formulating rules for shadow removal.
Innovation Solution
A system and method that statistically combines multiple cues like color, edge differences, and motion into a saliency map, using noise estimates to weight pixel-by-pixel and local neighborhood cues, allowing for more accurate foreground/background pixel decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pixel-based shadow removal algorithms are used, then processing speed is improved, but detection accuracy deteriorates due to ignoring textured objects and adjacent pixels being corrected in opposite directions
Solution Approach 1:
The patent combines multiple cues (color differences, edge differences, motion information) into a unified saliency map using statistical combination methods. This merging of multiple detection approaches allows the system to maintain processing efficiency while improving detection accuracy by considering multiple features simultaneously rather than relying on a single pixel-based algorithm that may correct adjacent pixels in opposite directions.
Solution Approach 2:
The patent introduces a saliency map as an intermediary representation that statistically combines information from multiple cues. This intermediary structure allows the system to integrate diverse information sources (color, edges, motion) in a coordinated manner, preventing the conflicting corrections that occur in pixel-based algorithms while maintaining computational efficiency.
2Measurement precision
If post-processing methods are used to detect shadows by looking for bounding edges, then detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent merges shadow detection with the primary object detection process by incorporating edge difference information as one of multiple cues in the saliency map. This integration allows edge-based shadow detection to occur simultaneously with object detection rather than as a separate post-processing step, thereby maintaining detection accuracy while reducing processing time.
Solution Approach 2:
The patent performs edge analysis and shadow detection as part of the preliminary cue extraction phase before final saliency map generation. By preparing edge difference information in advance as one of the input cues, the system avoids time-consuming post-processing operations while ensuring accurate shadow detection is incorporated into the main detection pipeline.
3Measurement precision
If texture comparison methods are used to determine when to retain or dismiss regions, then detection accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent combines texture information with color and motion cues in a unified statistical framework. By integrating texture comparison as one of multiple cues rather than a separate complex analysis step, the system achieves improved region retention decisions while managing computational complexity through the efficient statistical combination methodology.
Solution Approach 2:
The patent transforms texture comparison into a probabilistic parameter within the statistical combination framework. By expressing texture information as likelihood ratios or probability distributions that can be combined with other cues, the system simplifies the computational complexity of texture analysis while maintaining its contribution to detection accuracy.
Data Source
AI summary
Systems and methods for detecting visual objects by employing multiple cues include statistically combining information from multiple sources into a saliency map, wherein the information may include color, texture and/or motion in an image where an object is to be detected or background determined. The statistically combined information is thresholded to make decisions with respect to foreground/background pixels.