Object Detection and Tracking in Videos: Mean Shift and Fast Mean Shift Techniques, Slides of Applications of Computer Sciences

Download Object Detection and Tracking in Videos: Mean Shift and Fast Mean Shift Techniques and more Slides Applications of Computer Sciences in PDF only on Docsity!

Presentation layout

 Introduction  Scope  Applications  Summary of previous work  Mean shift technique  Fast Mean Shift Technique  Comparison  Time Line  Future work  Conclusion

Introduction

 The basic concept behind object detection in videos engrosses:

 Verification of the presence of an object.  Locating it in particular for recognition.

 Object’s temporal and spatial changes are measured during a sequence of images by solving the temporal correspondence problem

Scope (Cont-I)

 It finds its scope in fields like :

 Vision-based control systems

 Human-computer interfaces

 Medical imaging

 Augmented reality

 Robotics

Applications

 Some of the applications are

 Target tracking using video sensors  Assistance for handicapped people  Traffic monitoring  Drivers assistance ,Traffic signs recognition  Medical Imaging  Medical image quantization and analysis  Military applications  Missile tracking ,jet tracking  Video surveillance  Security systems

Summary of Previous Work

 Object detection

 Background Subtraction Method

 Object tracking

 Template matching technique  Performing a normalized cross-correlation between a template image and a new image

Experimental results of Background

Subtraction

These figures show the results of background subtraction algorithm i.e. frame differencing applied on a sequence of images

Complications while using Template Matching

 Complications arise specially in air borne object, due to significant change in target orientation in the very next frame which causes

 Loss of target by tracker

 Correlation value drops due to which template is not updated and in next 2 or 3 frames tracker completely looses the object.

Mean Shift Technique

 The Mean Shift algorithm is a non parametric technique used to locate density extrema or modes of a given distribution by an iterative procedure

 The Mean Shift algorithm is a mode-seeking process on the density function surface

Mean Shift Tracking (Cont-I)

 The difference image generated from given image sequence is contains large number of high-intensity peaks or modes.  Our principal objective is to find modes representing our desired object.  The search process is facilitated by information of expected scaling of object {H(yi),W(yi)}

Mean Shift Algorithm

 Step 1: The difference image intensity maximum is mapped to unit intensity

 Step 2: A sample set of n points X1 ….... Xn is defined by locating local maxima

 Finding local maxima

 Locating the global intensity maximum and adding it to the list of sample set  Resetting the difference image intensity  Repeating the maximum search of step (1) until the found maximum drops below a threshold T

Mean Shift Algorithm (Cont-II)

 Step 4: The convergence points of individual mean shift procedures are linked together forming the centers of detected clusters  The convergence trajectories leading to the cluster center defines the paths of the cluster.  Step 5: A bounding box around the cluster path gives a simple representation the region covered by the cluster.

Fast Mean Shift

 The convergence speed of Fast Mean Shift is faster than that of Mean Shift.  Basic motivation comes from the idea that in the case of linear operations (e.g. ), any invertible linear operation can be applied to f or g if its inverse is applied to the result.  As in convolution, if derivative operator is applied both to the image and the kernel the result must then be double integrated. f  g  ( f '^ * g ') 

Fast Mean Shift Computation

 The integral image at a location (x,y) contains the cumulative sum of pixels located to the left and above (x,y) including the pixel (x,y) :

' '

' ' int ,

x x y y

I x y I x y

 

I int ( , x y )  I int ( , x y  1)  I int ( x  1, y )  I x y ( , )  I int( x  1, y 1)

Fast Mean Shift Computation (Cont-I)

Using the integral image, the area sum of a rectangular region within the original image can be efficiently computed by the following step

Sarea  I A  I D  I B  IC

Sarea is the area sum within the rectangle