Professional Documents
Culture Documents
Behavior Analysis of Individuals
Behavior Analysis of Individuals
Behavior Analysis of Individuals
EE508
Dr Mahesh Kolekar
IIT Patna
Learning-based Behavior Analysis
xc, yc
Centroid
xi,yi
Distance of centroid from contour points
Distance of centroid from contour points
Kalman
Filter
Mean-Shift based Motion Feature Searching
Mean-shift method
to search for the
region that has
concentration of
motion
information.
Motion information
= current frame –
previous frame
Continuously Adaptive Mean Shift
1. Choose Initial location of search window and search window
size.
2. Compute mean location of search window and store the
zeroth moment.
3. Center the search window in the mean location computed in
the previous step.
4. Repeat 2 & 3 until mean location moves less than the preset
threshold.
5. Set search window size = function of zeroth moment in 2.
6. Repeat 2-5 until the mean location moves less than the preset
threshold.
Body Location
CAMSHIFT contd…
• Centroid of the moving part within the search window can
be calculated from M00, M10, M01
• Generally Mij= Ʃ Ʃ xi yj I(x,y)
• M00= Ʃ Ʃ I(x,y), M10= Ʃ Ʃ x I(x,y) M01= Ʃ Ʃ y I(x,y)
• I(x,y) value of difference image at (x,y).
• Centroid: xc = M10, yc = M01
M00 M00
History Motion Image Based Analysis
• HMI image is based on foreground image.
• A threshold is needed to transform foreground image into binary
image
1. Update the HMI template T by a foreground at different time stamp τ:
τ, I(x,y)=1
T(x,y)= 0, T(x,y)< τ-δ I difference image
T(x,y), else δtime window to be considered.
where Gy(x, y), Gx(x, y) are the sobel convolution results with the HMI
template in the x and y directions respectively.
Learning Based Behavioural
recognition
Multi-SVM based Bayesian
Network
Frame Work Analysis
Two approaches:
where .
denotes the dot product and the normal vector to the
hyperplane. The parameter determines the offset of the
hyperplane from the origin along the normal vector .
Maximum-margin hyperplane and margins for an SVM
trained with samples from two classes.
Samples on the margin are called the support vectors.
• For each action type, we will set up a SVM model to
give out the probability P( x|Hi )
• Recognizing