最近在看目标检测方面的文章,主要集中在颜色、纹理和分层,解决实际中遇到的问题。特将收集到的文章进行分享,并附有简单的介绍,文章在google scholar可以下到(也可以通过后面连接下载:http://download.csdn.net/detail/kezunhai/5283117).
1. Real-Time Moving Object Detection for Video Surveillance
对视频帧进行4×4的Patch,然后用DCT提出系数向量(Coefficient vectors),用这些进行背景建模。
2. 2008_ICPR_Detecting global motion patterns in complex videos
采用光流法来做,所以实时性达不到要求,(Flow vector)
3. A texture-based method for modeling the background and detecting moving objects
对每个像素在一个Circular区域内的LPB直方图进行建模。
4. Moving object segmentation by background subtraction and temporal analysis
该论文的算法可以好好的分析下,然后予以实现。(思路:提出了一个光照增益因子和辐射相似性的概念,利用了时域的帧差,并考虑了全局的相关概念)。
5. Object Detection Using Local Difference Patterns
Integrated Pixel-based and spatial-based approach,利用LDP(local Difference Pattern)进行了背景建模和实现前景检测。
6. Spatial-Temporal patches for night background modeling by subspace learning
提出Brick的概念,15×15×7bricks,利用subspace learning来进行目标检测。
7. Fusion of background estimation approaches for motion detection in non-static backgrounds
Fusing the results of short-term and long-term yields better performance, 其中long-term是通过Intensities Histogram的方法来处理的(邻近的N帧)
8. pami_Density-based multi-feature background subtraction with SVM
在该算法中,颜色、梯度和Haar-like特征综合起来。在Pixels-wise应用KDA(Kernel density approximation)来建立背景模型,在目标检测部分则根据提取的特征利用SVM来进行分割,具有很好的鲁棒性。
9. Statistical modeling of complex backgrounds for foreground object detection
提出了一种以能量、空间和时间为特征的背景模型的贝叶斯框架。(A Bayesian framework that incorporates spectral, spatial and temporal features to characterize the background appearance is proposed).根据统计的principal features,背景提取的Bayes decision rule is derived.
10. A framework for feature selection for background subtraction
提出了一个用于背景建模和减除的特征选择框架,对于每一个像素的特征组合采用RealBoost算法进行选择。通过一段时间的特征估计(KDE),建立特征池。
11. Towards Robust Object Detection: Integrated Background Modeling Based on Spatio-temporal Features.
It consists of three complementary approaches: pixel-level background modeling, region-level one and frame-level one. The pixel-level background model uses the probability den-sity function to approximate background model. The PDF is estimated non-parametrically by using Parzen density estimation. The region-level model is based on the evaluation of the local texture around each pixel while reducing the effects of variations in lighting. The frame-level model detects sudden, global changes of the the image brightness and estimates a present background image from input image referring to a background model image.
12. Learning a scene background model via classification
We consider a two-level approximation scheme that elegantly combines the bottom-up and top-down information for deriving a background model in real time. The key idea of our approach is simple and effective:如果一个分类器能够在一个学习模型中确定图像块的某一区域为背景,并通过局部块间一致性的全局验证来提高the quality of solution. The overall idea is to efficiently identify background blocks from each images frame through online classification, and to iteratively integrate these background blocks into a complete model.(in a progressive manner).
13. An Efficient Region-Based Background Subtraction Technique
Color histograms, texture information, and successive division of candidate rectangular image regions are utilized to model the background and detect the motion. The proposed method combines this principle and the Gaussian mixture background modeling to produce a new method which outperforms the classic Gaussian mixture background subtraction method. It has the advantages of filtering noise during image differentiation and providing a selectable level of detail for the contour of the moving shapes. 根据不同大小的矩阵块,计算统计直方图和矩阵块内的方差,用这些值结合GMM进行背景建模。
14. The medial feature detector-Stable regions from image boundaryiccv11
用于特征检测的,类似于(MSR、SIFT、SURF等)
15. Real-time background subtraction in dynamic scene_ICCV09
In this paper, we formulate the problem mathematically as minimizing a constrained risk functional motivated from the large margin principle(最大间隔). It is a generalization of the one class support vector machines (1-SVMs) to accommodate spatial interactions, which is further incorporated into an online learning framework to track temporal changes. As a result it yields a closed-form update formula, a central component of the proposed algorithm to enable prompt adaptation to spatial-temporal changes. 这篇文章将分类问题翻译成一个数学问题,采用SVM来进行处理。
16. Making background subtraction robust to sudden illumination changes
Proposed a statistical illumination model, which model the ratio of intensities between a stored background images and an input images in all three channels as a Gaussian Mixture Model that accounts for the fact that different parts of the scene can be affected in different ways.(作者提供了源代码,可以考虑弄来做实验)
17. Detected motion classification with a double-background and aNeighborhood-based difference
这里应用a double model来进行背景检测,分别为long-term background(主要应用了temporal Medium Filter)和short-term background。
18. Learning color and locality cues for moving object detection and segmentation.
(Feng Liu and Michael Gleicher. Learning color and locality cues for moving object detection and segmentation. IEEE CVPR 2009, Miami Beach, Florida, USA, June, 2009. pp. 320-327)
It presents an unsupervised algorithm to learn object color and locality cues from the sparse motion information. We first detect key frames with reliable motion cues and then estimate moving sub-objects based on these motion cues using a Markow Randow Field framework. From these sub-objects, we learn an appearance model as a Color Gaussian Mixture Model. To avoid the false classification of background pixels with similar color to the moving objects, the locations of these sub-objects are propagated to neighboring frames as locality cues. Finally, robuts moving object segmentation is achieved by combing these learned color and locality cues with motion cues in a MRF framework.
Step1: Detect key frames that contain motion cues that can reliably indicate at least some part of the moving object.
Step2: From the key frames, we estimate moving sub-objects based on motion cues using MRF model.
Step3: We then learn from these sub-objects a moving object color model characterized by a GMM.
Step4: To avoid false detection of background pixels with similar color to the moving objects, we propagate the location information of sub-objects to non-keyframes as locality cues
Step5: we extract the moving object by combining these learned color and locality cues with motion cues in MRF framework.
应用Sift算法提取帧内特征,然后在帧间进行匹配,并应用Ransac估计Homography。(Find Some time to Research it intensively)
19 Robust Object Detection Via Soft Cascade
We describe a method for training object detectors using a generalization of the cascade architecture, which results in a detection rate and speed comparable to that of the best published detectors while allowing for easier training and a detector with fewer features.
20. Multi-layer background subtraction based on color and texture.
In this paper, we propose a layer-based method to detect moving foreground objects from a video sequence taken under a complex environment by integrating advantages of both texture and color features. Firstly, we integrate a newly developed photometric invariant color measurement in the same framework to overcome the illumination of LBP features. Secondly, a flexible weight updating strategy for background models is proposed to more efficiently handle moving background objects. Thirdly, a simple layer-based background modeling/detecting strategy was developed to handle the background scene changes due to addition or removal of stationary objects.
经过统计发现,由于光照变化引起的像素值的变化主要分布在沿着RGB原点的轴线上。因此可以通过比较像素点与原点的夹角来计算两者的差异。
21. Efficient hierarchical method for background subtraction
In this paper, we propose a method that combines pixel-based and block-based approaches into a single framework. We show that efficient hierarchical backgrounds can be built by considering that these two approaches are complementary to each other. Contrast histogram is used as the features to describe each block(coarse-fine hierarchical method).
22. Using Partial Edge Contour Matches for Efficient Object Category Localization
In our contributions we focus on the partial matching of noisy edges to relax the constraints on local neighborhoods or on assigning entire edges as background disregarding local similarities. We formulate a category localization method which efficiently retrieves partial edge fragments that are similar to a single contour prototype. We introduces a self-containing descriptor for edges which enables partial matching and an efficient selection and aggregation of partial matches to identify and merge similar overlapping contours up to any length. A key benefit is that the longer the matches are, the more they are able to discriminate between background clutter and the object instance. In this way we lift standard figure/ground assignment to another level by providing local similarities for all edges in an image. We retrieve these partial contours and combine them directly in a similarity tensor and together with a clustering-based center voting step we hypothesize object locations.