Human Detection System Using Adaptive Background Modeling and HOG Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing human detection systems face challenges in accurately detecting humans in images, especially under unfavorable conditions such as distorted signals, background mix-ups, and varying human postures, with limited precision and high false positives, particularly in infrared images and changing backgrounds.
Innovation Solution
An adaptive learning-based system that combines Histogram of Oriented Gradients (HOG) and Haar-like wavelet features with adaptive background modeling for real-time human detection and counting, using a processor to evaluate the presence of humans in images and send detection decisions to actionable means for security and surveillance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional background modeling is used for human detection, then the system is simple to implement, but detection accuracy deteriorates under unfavorable conditions such as distorted signals, background mix-ups, and varying human postures
Solution Approach 1:
The patent combines multiple feature extraction techniques (HOG for human body detection and Haar wavelets for face detection) with adaptive background modeling to create a hybrid detection system. This merging of multiple approaches allows the system to maintain high detection accuracy across various conditions while managing complexity through integrated processing
2Measurement precision
If SVM based HOG features classifier is used with extensive training data, then detection precision improves, but training time and computational resources increase significantly
Solution Approach 1:
The system performs preliminary adaptive background modeling and feature extraction to pre-process images before applying the SVM classifier. By preparing the data in advance through adaptive background subtraction and HOG/Haar feature extraction, the system reduces the computational burden during actual detection while maintaining high precision
3Adaptability or versatility
If color information is used for background modeling, then detection performance improves for color images, but the system fails to work effectively for infrared images where color information is absent
Solution Approach 1:
The patent implements a universal adaptive background modeling approach that works with both color and grayscale (infrared) images. The system automatically adapts to the input image type by using intensity-based background subtraction for infrared images while maintaining the same algorithmic framework, making the detection system versatile across different imaging modalities
4Measurement precision
If Haar-like features are used for face detection, then face detection capability improves, but the threshold value for extracted features increases, reducing overall detection efficiency
Solution Approach 1:
The system segments the detection process into two stages: first using HOG features for human body detection to locate potential human regions, then applying Haar wavelet transformation specifically to those regions for face detection. This segmentation allows Haar-like features to be used efficiently only where needed, improving face detection capability while maintaining overall detection efficiency
Data Source
AI summary
A system for adaptive learning based human detection for channel input of captured human image signals, the system comprising: a sensor for tracking real-time images of an environment of interest; a feature extraction and classifiers generation processor for extracting a plurality of features and classifying the features associated with time-space descriptors of image comprising background modeling, Histogram of Oriented Gradients (HOG) and Haar like wavelet; a processor configured to process extracted feature classifiers associated with plurality of real-time images; combine the plurality of feature classifiers of time-space descriptors; evaluate a linear probability of human detection based on a predetermined threshold value of the feature classifiers in a time window having at least one image frame; a counter for counting the number of humans in the real-time images; and a transmission device configured to send the final human detection decision and number thereof to a storage device.