18-07-2012, 12:39 PM
PEDESTRIAN DETECTION AND TRACKING AT NIGHT TIME
PEDESTRIAN DETECTION.ppt (Size: 755.5 KB / Downloads: 95)
INTRODUCTION
Pedestrians – most vulnerable to accidents at NIGHT TIME
Pedestrian safety-a major concern in a highly populous country like India.
Solution-an integrated pedestrian detection and tracking system within the vehicle.
This method involves recognition of pedestrians in real time
Uses camera integrated to the vehicle for acquiring video of moving pedestrians
With the help of image processing , presence of humans and the probability of collision is calculated
The output of system can be used to generate appropriate warning for the driver or to perform automatic braking .
PEDESTRIAN DETECTION MODEL
SENSING UNIT
Different types of sensors are available for sensing the pedestrians. Mainly used are ultrasonic sensor, laser scanner, microwave radars and IR cameras.
Laser scanners and radars does not distinguish pedestrians from other obstacles.
Hence IR cameras are used as they provide rich information.
REGION OF INTEREST(ROI) GENERATION
IMAGE ACQUISITION
ROI means the regions that potentially contain the pedestrians.
Near infrared(NIR) cameras are used for capturing the video.
Here image of an object is detected based on the temperature of the object and the amount of heat it radiates.
It has several advantages than other type of cameras.
It is very cheap and highly practical.
CANDIDATE SELECTION
Involves selection of proper human object from the Segmented frame.
Candidate filtering:- It means the simple filtering of non human candidates by using some features such as height, width , aspect ratio etc
Candidate regenerating is a useful strategy for accuracy and high detection rate.
OBJECT CLASSIFICATION
In classification stage, features are extracted based on a classifier.
A cascaded classifier based on AdaBoost is implemented.
AdaBoost algorithm uses several weak classifiers to form a strong classifier
This gives efficiency as well as performance in small systems
TRACKING
In TRACKING stage, direction and speed of pedestrians are computed.
The TRACKING of pedestrian boundary is based on Kalman filtering
For tracking we takes : Directions X,Y and Z along with Width(W) and Height(H) – as state variables for Kalman filter
The filter gives the measure of variation of tracked object.
CONCLUSION
This paper has introduced a nighttime vision system for pedestrian detection and tracking from a moving vehicle. A cascade of three modules is involved in the system, and each module utilizes complementary visual features to successively distinguish the objects from the cluttered background.
In contrast to most other systems, in addition to the performance, the proposed system enjoys the advantage of low implementation cost, as only one NIR camera is required, and the core algorithm can perform as fast as the frame rate on a common PC platform.
The output of system can be used to generate appropriate warning for the driver or to perform automatic braking .