Far IR Pedestrian Detection Using DoG Noise Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pedestrian detection methods at night using infrared cameras face challenges such as image noise leading to candidate selection errors and long processing times due to motion analysis, especially when pedestrians are motionless, resulting in decreased accuracy.
Innovation Solution
A method utilizing a Difference of Gaussian (DoG) filter to process thermal images from far IR cameras, binarizing pixel values, and classifying pedestrian candidates based on behavioral characteristics to improve detection accuracy and reduce noise, with a classifier learning from motion patterns to enhance reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If gray level binarization is used to extract pedestrian candidates, then candidate selection is simplified, but image noise causes candidate selection errors
Solution Approach 1:
The patent segments the candidate selection process into multiple stages: initial gray level binarization to identify potential candidates, followed by variance-based filtering to remove noise-induced false candidates, and finally classifier-based verification. This multi-stage segmentation resolves the contradiction by maintaining simplicity in candidate identification while adding reliability through subsequent filtering steps.
Solution Approach 2:
The patent introduces variance calculation as an intermediary step between gray level binarization and final candidate selection. This intermediary mechanism computes the variance of gray levels around each candidate pixel, serving as a mediator that distinguishes true pedestrian candidates from noise-induced false candidates, thereby resolving the accuracy-simplicity contradiction.
2Reliability
If motion analysis is used to detect pedestrians, then detection capability is enhanced, but processing time increases significantly
Solution Approach 1:
The patent performs preliminary gray level binarization and variance-based candidate identification before applying motion analysis. By pre-processing the image to identify and mark potential pedestrian candidates, the system reduces the scope of subsequent motion analysis, thereby maintaining enhanced detection capability while significantly reducing processing time compared to analyzing the entire image area.
Solution Approach 2:
The patent applies motion analysis selectively only to regions identified as pedestrian candidates through gray level and variance analysis, rather than analyzing the entire image. This local application of motion analysis maintains detection reliability for actual pedestrians while reducing overall processing time by excluding non-candidate regions from computationally intensive motion analysis.
3Area of stationary object
If motion analysis is applied to the entire image area, then detection coverage is maximized, but processing time becomes excessively long
Solution Approach 1:
The patent segments the image processing into hierarchical stages: first identifying candidate regions through gray level binarization, then applying variance filtering, and finally performing motion analysis only on the reduced set of candidate regions. This segmentation maintains effective detection coverage by systematically identifying all potential pedestrians while reducing processing time through progressive filtering that eliminates non-candidate areas from computationally intensive analysis.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach effectively filters noise, improves pedestrian detection accuracy by classifying behavioral patterns, and reduces processing time, thereby enhancing night-time pedestrian detection and preventing accidents.
Implementation Method 1
receiving a thermal image of a pedestrian... using a far infrared ray (IR) camera... configured to project IR light
Data Source
AI summary
The present invention relates to a method for detecting a pedestrian based on a far infrared ray (IR) camera at night, which provides a method of receiving a thermal image of a pedestrian from a far IR camera, setting a candidate using a DoG filter having a robust characteristic against image noise, and accurately detecting the pedestrian using a classifier based on a behavioral characteristic of the pedestrian.