Introduction:
There are three key steps in video analysis: Detection of interesting moving objects, tracking of such this object from frame to frame, and analysis of object tracks to recognize their behavior. There,
Use of object tracking is pertinent in the tasks:
Motion-based regognition
Automatic surveillance
Video Indexing
Human-Computer Interaction
Traffic Monitoring
Vehicle Navigation
Tracking objects can be complex due:
Loss of information by detector;
Noise in images;
Complex object motion;
Nonrigid or articulated nature of objects;
Partial and full object occlusions;
Complex Object shapes;
Scene illumination changes;
Real-time processing requirements;
Some questions: Which object representation is suitable for tracking? Which image features shocould be used? How shocould the motion, appearance, and shape of the object be modeled?
2 object representation:
Points;
Primitive geometric shapes;
Object silhouette and contour;
Articulated shape models;
Skeletal models;
Probability densities of object appearance;
Templates;
Active appearance models;
Multiview appearance models;
3 Feature Selection for tracking;
Selecting the right features plays a critical role in tracking;
Color;
Edges;
Optical flow;
Texture;
Mostly features are chosen manually by the user depending on the application domain.
4 Object Detection
4.1 point Detectors
4.2 Background Subtraction
4.3 Segmentation
4.3.1 mean-shift Clustering
4.3.2 Image Segmentation Using Graph-cuts
4.3.3 active ipvs
4.4 supervised learning
Object detection can be passed med by learning different object views automatically from a set of examples by means of a supervised learning mechanism.
Selection of features plays an important role in the performance of the classification, hence, it is important to use a set of features that discriminate one class from the other.
Supervised Learning methods usually require a large collection of samples from each object class.
4.4.1 adaptive boosting
4.4.3 Support Vector Machines
5 Object Tracking
The aim of an object tracker is to generate the trajectid of an objet over time by location its position in every frame of the video.
Point Tracking
This approach requires an external mechanic to detect the objects in every frame.
Kernel tracking
Kernel refers to the object shape and appearance
This motion is usually in the form of a parametric transformation such as translation, rotation, and affine.
Silhouette tracking
Silhouettes are tracked by either shape matching or contour evolution.
5.1 point Tracking
5.1.1 deterministic methods for corresponsible
5.1.2 statistical methods for corresponsible
Statistical corresponsponmethods solve these tracking problems by taking the measurement and the model uncertainties into account during object state estimation.
5.1.2.1 single object State Estimation
Object state is not assumed to be a Gaussian, state estimation can be med using particle filters.
Kalman filters
Particle filters
Tracking multiple objects requires a joint solution of Data Association and state estimation problems.
5.1.2.2 multiobject Data Association and State Estimation
Joint probability data association filter
The major limitation of the jpdaf algorithm is its inability to handle new objects entering the field of view (FOV) or already tracked objects exiting the FOV.
Multiple hypothesis tracking
5.1.3 Discussion
One important issue in the context of point trackers is the handling of missing or noisy observations.
Point trackers are suitable for tacking very small objects which can be represented by a single point.
5.2 kernel tracking
Kernel tracking is typically saved by computing the motion of the object, which is represented by a primitive object region, from one frame to the next.
5.2.1 tracking using template and density-based appearance models
5.2.1.1 tracking single objects
Note that instead of templates, other object representations can be used for tracking, for instance, color histograms or mixture models can be computed by using the appearance of pixels inside the rectangular
Or ellipsoidal regions.
An online version of the EM algorithm is used to learn the parameters of this three compoente mixture.
Another approach to track a region defined by a primitive shape is to compute its translation by use of an optical flow method.
5.2.2 tracking using multiview appearance models.
In the previous tracking methods, the appearance models, that is, histograms, templates etc. are usually generated online. Thus these models represent the information gathered about the object from the most
Recent observations. The objects may appear different from different views, and if the object view changes dramatically during tracking, the appearance model may no longer be valid, and the object track might be lost.
5.2.3 Discussion
The main goal of the trackers in this category is to estimate the object motion.
5.3 silhouette tracking
Object may have complex shapes, for example, hands, Head, and shoulders that can not be well described by simple geometric shapes. silhouette based methods provide an accurate shape description for these
Objects.
5.3.1 Shape Matching
Shape Matching can be performed med similar to tracking based on template matching
Some approach to match shapes is to find corresponding silhouettes detected in two consecutive frames.
The main difference between silhouette matching and point matching is the object representations and the object models used.
Silhouette detection is usually carried out by background subtraction.
Tracking silhouettes can be saved by computing the flow vectors for each pixel inside the Silhouette such that the flow that is dominant over the entire silhouette is used to generate the Silhouette trajec.pdf.
5.3.2 Contour Tracking
Iteratively evolve an initial Contour in the previous frame to its new position in the current frame.
5.3.2.1 tracking using state space models.
5.3.2.2 tracking by direct minimization of contour energy functional.
5.3.3 Discussion
Silhouette racking is employed when tracking of the complete region of an object is required.
Generally the region-based approaches are more resilient to noise.
Occlusion handling is another important aspect of silhouette tracking methods.
Another important issue related to silhouette trackers is their capability for dealing with object split and merge.