Scribd
Upload
50 views31 pages

Guide Prof. P.J Engineer Co-Guide Prof. M.C Patel: Prepared by Parthiv Bharti P09 EC 916

Kalman filtering can be used for object tracking by representing objects as points and tracking their motion over time. The document discusses various object representations like points, shapes, silhouettes and graphs that can be used for tracking. It also covers feature selection, object detection methods like blobs and interest point detection, and tracking algorithms like mean shift, kernel and silhouette tracking. Resolving occlusions and tracking across multiple cameras are identified as open challenges.

Uploaded by

parthiv bharti
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
50 views31 pages

Guide Prof. P.J Engineer Co-Guide Prof. M.C Patel: Prepared by Parthiv Bharti P09 EC 916

Kalman filtering can be used for object tracking by representing objects as points and tracking their motion over time. The document discusses various object representations like points, shapes, silhouettes and graphs that can be used for tracking. It also covers feature selection, object detection methods like blobs and interest point detection, and tracking algorithms like mean shift, kernel and silhouette tracking. Resolving occlusions and tracking across multiple cameras are identified as open challenges.

Uploaded by

parthiv bharti
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 31

HDL Implementation of

Object Tracking through


Kalman Filtering

Guide Co-Guide
Prof. P.J Engineer Prof. M.C Patel
Prepared by
Parthiv Bharti
P09 EC 916
Overview
2
Applications of Object Tracking
3
Complexities Involved
4

Complex object shapes and motions

Non Rigid or articulated nature of Objects

Partial or full object occlusions

Changes in illumination levels

Real Time Processing requirements


Assumptions that Simplify…..
5

Object motion is smooth

No abrupt changes in shape

Constant velocity and/or acceleration in motion

Prior knowledge about shape and size

Appearance and illumination do not vary


Status
6
Object Representation
7

Point Representation

Primitive Geometric Shapes

Object Silhouette

Skeletal Models

Articulated Shape Models


Appearance Representations
8

Templates
Simple geometric
Characterize an image region
shapes/silhouettes by itposes
whose statistics.
do
If the statistics
not vary differ from background, they
considerably
should enable tracking.
Choosing Representations
9

Point representations for very small objects

Primitive shapes are results of approximations

For complex objects Silhouette scores


Feature Selection
10

Colour Influenced by illumination


variation
In general, the most desirable property of
Edges Less is
a visual feature sensitive to
its uniqueness so that
the objects illumination
can be easily changes
distinguished in
Texture
the feature space

It is measure of intensity
variation of a surface which
quantifies properties such as
smoothness and regularity.
Choosing Features
11

Choice is application dependent

Colour is the most popular one

Automatic feature selection is the future

Combination of various features improves


performance
Status
12
Object Detection
13

Blobs

In every frame

Use of temporal information reduces false


detection
Point Detection
14
find interest points in images which have an expressive
texture in their respective localities
Moravec’s Interest Operator

Harris Interest Operator

Kanade-Lucas-Tomasi [KLT] Detector

Scale Invariant Feature Transform [SIFT]


Detector
Moravec’s Operator
15

Compute variation of image intensities in a 4x4


patch in horizontal, vertical, diagonal, and anti-
diagonal directions

Select the minimum of four variations to


represent a window

An interesting point is a local maximum in


12x12 patch
Harris Detector
16

Compute the first order derivates in X and Y


directions

Second moment matrix is evaluated for each


pixel in a small neighbourhood

An interest point is identfied using


determinant and trace of this matrix
SIFT Detector
17

SIFT feature descriptor is invariant to scale,


orientation, and affine distortion, and partially
invariant to illumination changes
Produce image with different scales
Convolve each with Gaussian kernel
The differences between adjacent scales of
convolved images are calculated
Candidate keypoints are local maxima and minima
of the difference
Segmentation and Background
Subtraction
18

Segmentation partitions the image into perceptually


similar regions
Any significant change in an image region from
the background model signifies a moving object
Status
19
Object Tracking
20

Generate the trajectory for an object over time by


locating its position in every frame of the video

Real Time Constraint

Use only a small portion of the model space to


reduce the computational burden
Point Tracking
21

Point
Objects represented by points Tracking

Point matching is done

External Point detecting mechanism required


Possible Paths

Eg :Kalman Filter, Particle Filter

Constraints upon Movement


Kernel Tracking
22

Kernel = object shape and appearance

E.g. A rectangular template or an elliptical shape


with an associated histogram

Objects are tracked by computing the motion


(parametric transformation such as translation,
rotation, and affine) of the kernel in consecutive
frames
Mean Shift Tracking
23

Objects are represented by their color-histograms

Iteratively compares the histogram of the


original object in the current frame and that of
candidate regions in the next frame of image.

The aim is to find maximum correlation between the


two histograms
Mean Shift Tracking
24
Silhouette Tracking
25

Difficulty
Uses the information encoded inside the object
region (appearance density and shape models)
Object Split and MergeShape Matching
Silhouettes are
tracked by Contour Evolution
Graph-based Tracking
26

Graphs offer a way to represent the structure in a


rich and compact manner

Node attributes: size, average color, position

Edges specify the spatial relationships(adjacency,


border) between the nodes

In this way, each image of a sequence is segmented


and represented as a region adjacency graph

Object tracking becomes a graph-matching problem


Status
27
Issues
28

Resolving Occlusions

Multiple Camera Tracking


Future Directions
29

Development of roboust trackers in Real Time


Environment

Tracking in unconstrained videos

Multiple People Tracking

Combining audio along with video to


overcome occlusion
Status
30
31

You might also like