Você está na página 1de 19

DEPARTMENT OF ELECTRONICS & COMMUNICATION ENGINEERING VIGNANA BHARATHI INSTITUTE OF TECHNOLOGY (Affiliated to Jawaharlal Nehru Technological University ) Aushapur,

Ghatkesar,R.R.DIST

Moving Object Detection Based on Multiple Quadrant Histogram

Under the guidance of Mr. Kattaswamy By M.Srikanth (11P61D5510

Contents

Introduction Moving Object Detection Project Abstract Development and Target System Reference Paper Detection Algorithm Project Flow chart Development System

Event Trigger
Extracting Object Information Applications Future Work Conclusions

Introduction Moving Object Detection


Why Moving Object Detection

The core of security and business intelligence surveillance video application is the Moving Object Detection. The information derived from the process can be used to alert security camera monitoring staff to potential trespass or rule violation in the sensitive areas. In case of business intelligence the information can be used to gather positive statistical information relating to customer behavior.

Project Abstract
Problem Statement - Low Processing Power, Background less Moving Object Detection As the complexity of the video analytics processing increases in embedded system, more will be the usage of efficient methods to detect the moving objects. Most of the video surveillance method use background model and the correctness of the surveillance depends on the background and adaptive update to background. However this method take sufficient processing bandwidth and embedding in to low cost solution becomes not feasible. Alternately Histogram based detection is fairly efficient method to detect moving object. This method can be combined with multiple quadrant event trigger to quickly detect object.

So this project aims to detect the moving object using technique which does not require background model and derive a effective mechanism which can be embedded in ARM Board
Development Platform: MATLAB Target Platform : TI ARM Processor

OS
Activity Application

: Linux
: Image processing Algorithm : Video Surveillance
4

Development and Target System


Development System Target System

Hardware Platform

RS232

JTAG

USB Camera PC with MATLAB

USB Camera ARM Processor Ethernet

Microsoft Webcam

Image

acquisition

SDRAM

NAND Flash

30 frames/minute

Algorithm development

PC with MATLAB

Reference Paper Detection Algorithm


Based on single Gaussian Distribution
The color of every pixels are shown by Gaussian distribution model Color and covariance differ, then pixel is

considered as foreground pixel


After foreground detection, use average value and covariance matrix of Gaussian to update Use inter-frame difference method and find the

difference between two adjacent frames in video


sequence Update background model Obtain moving object by processing difference between current frame and updated background model

Project Flow chart

PAD/ Quadrant Four Quadrant

Project Flow chart

Development System
The implementation of the moving object identification is planned in two phases. In Phase 1, the moving object detection algorithm is developed in MATLAB.In Phase 2, this algorithm is ported to target Platform.
Status S.No 1 2 3 4

Activities MATLAB Setup Study of Image Processing Commands Image Acquisition using MATLAB Image Acquisition Tool Box Displaying Video Frames in MATLAB

Status Completed Completed Completed Completed

Beagle Board - xM

Laptop MATLAB Webcam

Output File Path Frame Width and height No of Frames Traffic.avi C:\Users\srikanthm\Desktop 160 &120 61

Development System

Event Trigger
The implementation of the Event Trigger is done using MATLAB Program. The detector is a Pad which covers an image region with dimensions 40x40 with starting Co-ordinate 40x60 The first step is construction of the threshold pad temporal difference image Dt

Where Pt Pad image at time t - Pixel Difference Threshold For the traffic.avi file, the pad temporal difference image is constructed between frame1 and frame 61 where the difference exists The mean difference across the pad is calculated

Where N- is the number of pixels in the pad image dt- mean differences threshold

The trigger occurs when increasing.

rises through a mean difference

and the mean differences are

Extracting Object Information


In absence of a trigger, a normalized 8 bin probably background grey scale histogram of the pad image at time t. The histogram is used to recursively update a time averaged probably background histogram as given below

Where - pad image histogram typically of the order 0.2

Event Trigger Pad Object

Frame 1 t0

Frame 2 t1

Frame 31 t30

Frame 32 t31

Frame 60 t59

Frame 61 t60

Probably background Histogram

Average

Difference Histogram

Event Trigger Pad Object

Frame 1 t0

Frame 2 t1

Frame 31 t30

Frame 32 t31

Frame 60 t59

Frame 61 t60

Probably background Histogram

Average

Difference
12

Difference Masked Pad Image Mp

Pad

Pt(x,y)
Frame 1 t0 Frame 2 t1

Object

Frame 31 t30

Frame 32 t31

Frame 60 t59

Frame 61 t60

Dt(x,y)

x
Mp(x,y)

This contains objects, shadows, ghosts and zero values associated with static region
13

Difference Masked Pad Image Mp

Mp(x,y)

Dirac function of Compare gray value with bin boundary

U=1 U=2

U=8

14

U=1 U=2

U=8

+
U=1 U=2

U=8
15

Applications

Video Surveillance
No processing time waste in updating background Reality and Visual Effects Medical Imaging

the security monitoring domain

16

Future Work

A multiple pad can make a tripwire type arrangement which can be effectively used for Tracker initialization. The existing method can be enhanced with additional tracker initialization and be used effectively in security alerts.
This embedded camera system may include short range wireless alert message features which can alert people around radius of 100m. The future scope includes making a independent, smart video analytics device embedded in to camera itself which has facility to communicate to surrounding intelligent devices too.

17

Conclusions
We have presented a computationally economic approach to detection of event as well as object detection and identification. It is one of a number of approaches developed to use spare processing capacity for embedded analytics in intelligent cameras. This method significant differs from other methods by efficient way of event trigger and histogram based detection.
It has potential for development in contexts such as Railway Platform density indication, or as a object identification in situations such as remote video surveillance. The method does not demand the development of a full background image or classifier training. It works with moderate quality monochrome footage and can be used in a range of contexts.

18

Thank You

19

Você também pode gostar