Incremental Object Detection Using Dual-Threshold Local Binary Pattern Operators
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer vision systems for object detection in mobile devices face high power consumption due to continuous image processing and feature extraction, leading to reduced battery life, especially in always-on applications.
Innovation Solution
The implementation of incremental feature extraction using a multi-stage process with band-pass feature extractors like local binary pattern (LBP) operators, which selectively extract features in stages and make partial decisions, reducing unnecessary processing and power consumption by only extracting features necessary for object detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If continuous image processing and feature extraction is performed to ensure accurate object detection, then detection reliability is improved, but power consumption increases
Solution Approach 1:
The patent divides the feature extraction process into multiple stages (first stage and second stage), where each stage extracts features at different resolutions or detail levels. This segmentation allows the system to perform coarse feature extraction first, then selectively refine only those regions that require higher detail, thereby maintaining detection accuracy while reducing overall computational load and power consumption.
Solution Approach 2:
The patent implements partial feature extraction by only processing certain image regions or features in the second stage based on results from the first stage. Instead of extracting all features at full resolution, the system performs extraction selectively on relevant portions, achieving sufficient detection reliability with reduced processing effort and lower power consumption.
2Reliability
If all features are extracted at once to ensure complete object analysis, then detection completeness is improved, but processing time increases
Solution Approach 1:
The patent segments the feature extraction process into multiple stages, where the first stage performs initial feature extraction and the second stage performs refined extraction on selected regions. This multi-stage segmentation enables the system to complete the detection process faster by avoiding unnecessary full-resolution feature extraction across the entire image, thus reducing processing time while maintaining completeness.
Solution Approach 2:
The patent performs preliminary feature extraction in the first stage to identify regions of interest or potential object locations. Based on these preliminary results, the second stage focuses computational resources only on relevant regions, eliminating the need to process entire images at full detail. This preliminary action significantly reduces processing time while ensuring complete analysis of important features.
3Measurement precision
If high-resolution feature extraction is performed to improve detection precision, then measurement precision is improved, but use of energy increases
Solution Approach 1:
The patent applies local quality by performing high-precision feature extraction only in specific regions of interest identified in the first stage, rather than uniformly across the entire image. This localized approach ensures that measurement precision is maintained where needed while avoiding unnecessary high-resolution processing in other areas, thereby reducing overall energy consumption.
Solution Approach 2:
The patent implements partial high-resolution feature extraction by selectively applying detailed processing only to regions that require it, based on results from the first stage. This partial action approach maintains measurement precision for critical features while avoiding excessive processing energy expenditure on regions that do not require high-resolution analysis.
Data Source
Figure 1
Figure 2
Figure 3A~3D
AI summary
Methods, systems, computer-readable media, and apparatuses for incremental object detection using a staged process and a band-pass feature extractor are presented. At each stage of the staged process, a different band of features from a plurality of bands of features in image data can be extracted using a dual-threshold local binary pattern operator, and compared with features of a target object within the band for a partial decision. The staged process exits if a rejection decision is made at any stage of the staged process. If no rejection decision is made in each stage of the staged process, the target object is detected. Features extracted at each stage may be from a different image for some applications.