565 GPU based human detection for Video Surveillance Applications (original) (raw)
Related papers
Histograms of Oriented Gradients for Human Detection
We study the question of feature sets for robust visual object recognition, adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of Histograms of Oriented Gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping descriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds. Receiver Operating Characteristics (ROC's) but allow small
We study the question of feature sets for robust visual object recognition, adopting linear SVM based human detection as a test case. After reviewing existing edge and gradient based descriptors, we show experimentally that grids of Histograms of Oriented Gradient (HOG) descriptors significantly outperform existing feature sets for human detection. We study the influence of each stage of the computation on performance, concluding that fine-scale gradients, fine orientation binning, relatively coarse spatial binning, and high-quality local contrast normalization in overlapping de-scriptor blocks are all important for good results. The new approach gives near-perfect separation on the original MIT pedestrian database, so we introduce a more challenging dataset containing over 1800 annotated human images with a large range of pose variations and backgrounds.
Enhancing Real-time Human Detection based on Histograms of Oriented Gradients
In this paper we propose a human detection framework based on an enhanced version of Histogram of Oriented Gradients (HOG) features. These feature descriptors are computed with the help of a precalculated histogram of square-blocks. This novel method outperforms the integral of oriented histograms allowing the calculation of a single feature four times faster. Using Adaboost for HOG feature selection and Support Vector Machine as weak classifier, we build up a real-time human classifier with an excellent detection rate.
Accelerated Histogram of Oriented Gradients for Human Detection
Dutse Journal of Pure and Applied Sciences
Histogram of Oriented Gradients (HOG) is an object detection algorithm used to detect people from an image. It involves features extraction called ‘HOG descriptor’ which are used to identify a person in the image. Several operations are involved in the feature extraction process. Hence performing numerous computations in order to obtain HOG descriptors takes some considerable amount of time. This slow computation speed limits HOG’s application in real-time systems. This paper investigates HOG with a view to improve its speed, modify the feature computation process to develop a faster version of HOG and finally evaluate against existing HOG. The technique of asymptotic notation in particular Big-O notation was applied to each stage of HOG and the complexity for the binning stage was modified. This results in a HOG version with a reduced complexity from n4 to n2 thereby having an improved speed as compared to the original HOG.
Fast human detection with cascaded ensembles on the GPU
2010 IEEE Intelligent Vehicles Symposium, 2010
We investigate a fast pedestrian localization framework that integrates the cascade-of-rejectors approach with the Histograms of Oriented Gradients (HoG) features on a data parallel architecture. The salient features of humans are captured by HoG blocks of variable sizes and locations which are chosen by the AdaBoost algorithm from a large set of possible blocks. We use the integral image representation for histogram computation and a rejection cascade in a slidingwindows manner, both of which can be implemented in a data parallel fashion. Utilizing the NVIDIA CUDA framework to realize this method on a Graphics Processing Unit (GPU), we report a speed up by a factor of 13 over our CPU implementation. For a 1280×960 image our parallel technique attains a processing speed of 2.5 to 8 frames per second depending on the image scanning density, which is similar to the recent GPU implementation of the original HoG algorithm in .
2011
Pedestrian Detection is of interest in many computer vision applications such as intelligent transportation systems and human-robot interaction; among the existing methods, the combination of shape feature (i.e. Histogram of Oriented Gradients (HOG)) and texture features (i.e. Local Binary Pattern (LBP)) has shown promising results in detection accuracy, but it is limited due to computation cost. In this paper, we introduce a new pedestrian detection algorithm with fast computation of these features on GPU. We propose a robust and rapid pedestrian detector by combining the HOG with LBP, as the feature set and corresponding Support Vector Machine (SVM) classifiers. Also, we use the integral image method and an efficient parallel implementation to reduce detection time. We can achieve a more than 10× speed up, and 7% increase in detection rate.
Boosting Histograms of Oriented Gradients for Human Detection
In this paper we propose a human detection framework based on an enhanced version of Histogram of Oriented Gradients (HOG) features. These feature descriptors are computed with the help of a precalculated histogram of squareblocks. This novel method outperforms the integral of oriented histograms allowing the calculation of a single feature four times faster. Using Adaboost for HOG feature selection and Support Vector Machine as weak classifier, we build up a fast human classifier with an excellent detection rate.
Human detection based on integral Histograms of Oriented Gradients and SVM
2011 International Conference on Communications, Computing and Control Applications (CCCA), 2011
This paper presents a method for human detection in video sequence. The Histogram of Oriented Gradients (HOG) descriptors show experimentally significantly out-performs existing feature sets for human detection. Because of HOG computation influence on performance, we finally choose a more better HOG descriptor to extract human feature from visible spectrum images based on OpenCv and MS VC++. We realized an image descriptor based on Integral Histograms of Oriented Gradients (HOG), associated with a Support Vector Machine (SVM) classifier and evaluate its efficiency.
Significant HOG-Histogram of Oriented Gradient Feature Selection for Human Detection
International Journal of Computer Applications, 2015
Detecting human efficiently is an important field of research and has many applications such as intelligent vehicle, robotics and video surveillance. Histogram of Oriented Gradient (HOG) is one of the eminent algorithms for human shape detection. HOG features are extracted from all location of a dense grid on an image region and use linear Support Vector Machine (SVM) to classify the combined features. Although HOG gives an accurate description of the contour of human body, it requires a large computational time. We studied the fundamental idea and consider features that have high percentage to contain edge. In this proposed method we used difference of Gaussian to obtain the edge percentage of each feature. Then a threshold is used to remove features with low edge percentage. Selected features then classified using linear SVM. Experiments on INRIA dataset demonstrate that the proposed method not only reduce the dimension of the HOG features but also outperforms.
Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications, 2016
Vision based human detection continuously attracts research interest since it is a topic of practical significance. The well-established Histogram of Oriented Gradients (HOG) human detector, though regarded as a reference for human detection, still suffers from the typical problem of the trade-off between precision and recall, relying on the threshold of its classifiers. In this paper, we propose a human detection system which can provide both good precision and recall without the need for adjusting the classification thresholds. Our strategy is to combine the HOG detector with a body part model in order to eliminate the false detections that do not match the human silhouette (body) model. For this purpose, a probabilistic model of the human body is learned to describe the relative position between the distinctive body parts. A HOG detection would be retained if the body parts can be detected in the confidence areas provided by the learned body model. Moreover, the body parts detectors are boosted cascade classifier learned with the Haar, HOG or LBP features. The multi-modal feature representation of the different human body parts is more robust against variations in human appearances. Experiment results on the INRIA data sets show that our human detector achieves a precision of 70% at a recall of 50%, which cannot be achieved by the HOG detector under any parameter settings.