You are on page 1of 15

Real-time Multiple from An Analysis

and

Sagar Medikeri, Shashank Pujar Dept of Electronics & Communications, B V Bhoomaraddi College of Engineering & Technology, Hubli, India.

Surveillance - call for human attention only when necessary Assistive Technology Human-Computer Interface Law Enforcement, Defense Pedestrian Detection for Cars

3D Model Recovery requires Stereo vision, sensors Appearance-based Model 2D features Example: Haar Classifier Motion-based Model Motion characteristics

Run Left Walk Left Stand Still Walk Right Run Right

OpenCV library contained Haar Classifiers haarcascade_fullbody.xml only detects persons oriented towards the camera. highly computation intensive, very time-consuming.

Video source: www.istockphoto.com

Input Video

Bounding Box & Activity Tagging

Contouring Difference Silhouette & MHI

Motion Vector Calculation

Median Filtering & Dilation Noise Removal


Video source: www.istockphoto.com

Sources of Noise Blowing of leaves Ripples on water surface appears as Salt-n-pepper noise Median Filtering Dilation clubs all close movements together

Find Contours and draw bounding boxes Compute Centers of boxes Find Euclidean distance for all combinations x = [x(t) x(t1)] y = [y(t) y(t1)] d = [(x)2 + (y)2] Contour pair producing smallest difference are linked and x is preserved

Disp = x (t)
t = -L

By heuristics, L=6

Frames 2000 1800 1600 1400 1200 1000 800 600 400 200 0

1900 1812 1370 1260

Ground Truth Detected 87 65 50 10

100 64

87 8

Statistical Modelling Adaptive Thresholding to distinguish between Walking & Running states Include more states in Activity Set Sitting, Fidgeting Haar Training

Huang T.S. Hong P., Turk M. Gesture modeling and recognition using finite state machines. Proceedings of the Fourth IEEE International Conference and Gesture Recognition, Grenoble, France, 2000. M. E. Pollack and L. E. B. et al. Autominder: an intelligent cognitive orthotic system for people with memory impairment. Robotics and Autonomous Systems, 2003. Paul E. Rybski and Manuela M. Veloso. Robust realtime human activity recognition from tracked face displacements. September 2005.

You might also like