|
|
| Object Detection
1. Overview
Given an image, object detection is to determine whether or not the
specified object is present, and, if present, determine the locations and sizes
of each objects". The research for object detection and recognition is focusing on
1) Representation: How to represent an object,
2) Learning: Machine Learning algorithms to learn the common property of a class of objects
3) Recognition: Identify the object in an image using models learned from 2).
Depending on how we represent an object, we can divide the object detection methods as Global representation and Part-Based representation. Depending on the machine learning algorithms, we can divide all object detection methods as Generative Methods and Discriminative Methods. Usually the recognition methods vary with respect to object detection applications. Major object detection algorithms, therefore, can be roughly classified as the following figure.
Figure 1. Overview of Object Detection Methods
2. Global Representation with Discriminative Learning
3. Part-based Representation with Generative Learning
Deformable
objects or articulate objects can be described as a combination of
their parts, so that these objects can be properly described using
graphic models.
Graph StructuresDifferent
objects, due to their distinct structure and texture properties, might
results in different graph models. for example, Composition model, Constellation Model, and Pictorial Model. Optimization of computations Computational
speed is another issue in the part-based method object detection
algorithms. Simple graph model usually results in faster computation
while complicated graph models need optimization in the computations.
Dynamic program, belief propagation algorithms are already applied to
accelerate the computation.
Defintion
Articulated objects are usually referred to "A multi-body system
composed of at least two rigid components and at most six independent
degrees of freedom between any two components". For example, human
bodies can be regarded as articulated objects where the arms, legs,
head and torch are rigid objects. Other examples of articulated objects
include human hands and animals. Issues
Despite its wide applications, the research on articulated object
detection is still limited to experimental systems. That is, there are
still no reliable practical commercial systems in the markets because
of the difficulty of detecting articulated objects. The difficulty lies
in two aspects: the shape variance and the self-occlusion. Because the
large number of degrees of freedom of articulated objects, it is hard
to build a shape model to model all possible shapes of articulated
objects, although some researchers did build such models. The other
factor is the self-occlusion of articulated objects. Approaches
Previous articulated object detecting systems, in order to deal with
the large shape variance, either take the "pose-based" approaches or
part-based approaches. 5. Part-Based Representation with Discriminative Learning
6. References[1]J.M.Coughlan, D.
Snow, C. English, and A.L.Yuille, "Efficient Deformable Template,
Detection and Localization without User Initialization". Computer
Vision and Image Understanding. 78, pp 303-319, 2000
[2]
P.Felzenswalb, "Representation and Detection of Deformable Shapes".
IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.
27, No. 2. 2005
[3] J.M.Coughlan, , and S. Ferreira, "Finding
Deformable Shapes using Loopy Belief Propoagation". In Proceedings
European Conference of Computer Vision.. 2002.
[4] J.M. Coughlan, ,
and H. Shen, "Shape Matching with Belief Propagation: Using Dynamic
Quantization to Accomodate Occlusion and Clutter". In GMBV. 2004
[5]Dong-Qing Zhang and Shih-Fu Chang, A Generative-Discriminative Hybrid Method for Multi-View Object Detection, CVPR 2006 [6]D.Crandall and P.Felzenszwalb. Spatial priors for part-based recognition using statistical models, CVPR 2005 [7]A.Holub and P.Perona. A discriminative framework for modeling object class. CVPR 2005 [8]R.Fergus, P.Perona, and A.Zisserman. Object class recognition by unsupervised scale-invariant learning, CVPR 2003 [9]M.Weber, M.Welling, and P.Perona, towards automatic discovery of object categories | |
|