object selection using freehand sketches

FIRST PROJECT“Motion detection in urban traffic”
Group No.11
Group MembersShilpa Sarawagi (Y08uc111) &
Disha Ajmera(Y08uc051)
InputA video sequence of urban
traffic.
OutputIdentifying moving objects
from that video sequence
which differ significantly from
the background. For eg,
vehicles , pedestrians ,etc.
Approaches that may be used1)Division of frames at a specific rate.
2)Background separation.
3)Connected component analysis
4) Metamorphological opening-Erosion
and dilation.
Challenges associated1)avoid detecting non-stationary background
objects such as swinging leaves, rain etc using
thresholding.
2)Robust against changes in illumination.
3)Finally, the constructed background model
should react quickly to changes in background
such as starting and stopping of vehicles.
Tentative Deadlines1)Division of frames by some technique.
(by 4-5 nov)
2)background separation.
(by 11-12 nov)
3)Connected component analysis.
(by 17-18 nov)
4) erosion and dilation.
(by 22-23 nov)
second PROJECT-
Input and Expected Output

Input : The inputs to the algorithm are
the image containing the object to be
selected and the freehand sketches drawn
by a human user over the image.

Expected output : The algorithm allows
selection of image objects with complex
boundaries using only roughly drawn
simple sketches.
Approaches that may be used1)Sketch processing after taking the
freehand sketch.
2)The initial selection and the boundary
triangles are then processed for local alpha
estimation algorithm (implemented to track
boundary)
3)Segmentation.
Challenges associated1)Object may not have well defined boundaries.
2)There may be overlapping objects.
Tentative Deadlines1)Sketch processing after taking the freehand sketch.
(6-7 nov)
2)The initial selection and the boundary triangles are
then processed for local alpha estimation
algorithm (implemented to track boundary)
(15-16 nov)
3)Segmentation.
(22-23 nov)
Third project
Input and Expected Output

Input : An aerial image containing the
coastline (which has to be tracked) or the
boundary to be followed like a road
surrounded by forests.

Expected Output : An image representing
the boundary or a binary image
differentiating the road from its
surroundings (such as forests).
Approaches and Challenges :

Approaches : Estimating the boundary of a
specific region of interest from an aerial
image includes two phases:
 First, a segmentation algorithm labels each pixel in
the scene as either belonging to the target region or
not, hence clustering the same.
 And then, a curve is fit through the connected group
of edge elements in the binary labeled image
corresponding to the desired boundary.

Challenges associated :
 The sand and water do not have a well-defined
boundary (in case of detecting a coastline).
 The algorithm should be robust to the illumination
changes.
Tentative Timeline:

Segmentation Phase (hue-based clustering
or texture based)
(by 2nd-3rd Nov.)

Contour Detection Phase :
 Edge Detection
(by 11th-12th Nov.)
 Temporal boundary tracking or per-frame boundary
detection
(by 17th-18th Nov.)

Final Presentation
(by 25th Nov.)