Naveen Presentation

YS
by yamika sharma0 views·Updated Jan 19, 2017
Powered by AI

Need to present on this topic?

Create your own polished presentation in minutes. Just describe what you need.

Free to try, no credit card required

Transcript

By

-Naveen Kumar Badam

Contents

  • INTRODUCTION
  • ARCHITECTURE OF THE PROPOSED MODEL
  • MODULES INVOLVED IN THE MODEL
  • FUTURE WORKS
  • CONCLUSION

introduction

  • A Video Surveillance System that detects Abandoned packages automatically
  • In this system multiple cameras locate objects in space and time despite occlusions and distracting lighting effects observed by subsets of cameras.
  • The system by describing the modules for camera view segmentation , object classification ,view object association ,3d object tracking and finally detection of the event of package being abandoned.

Features of The System

  • An abandoned package is any stationary package away from anyone considered responsible for it.
  • Is the object of interest a person, a displaced background object, or a package carried in?
  • How long has it been present?
  • Where is the package?
  • To whom does the package belong?
  • Is the person who brought it still nearby?

The Current situation

  • Our approach differs in at least two major ways from previously reported work.
  • First, we analyze relationships between objects. The owner of each abandoned object is determined and tracked using distance and time constraints through a multi-state model.
  • Second, we have exploited multiple cameras with overlapping fields of view to cope with occlusions of various types, and have empirically observed this to be essential in realistic situations.

Architecture of the model

Architecture overview

Processing Overview

An overview of the architecture of our approach, Figure (a), shows that video from each camera is separately processed before a combined processing phase. The percamera view processing Figure (b) outputs foreground regions (blobs) that are timestamped and registered in 3- space, by performing the following steps:

  • Lens spatial-distortion correction using intrinsic camera parameters from calibration.
  • Cross channel color correction , noise filtering (median, gaussian).
  • Foreground segmentation using an adaptive background model.
  • Region processing to combine spatially local regions for the same object.
  • Map from 2D screen coordinates into the common 3D coordinate system, using a projection matrix determined during offline calibration and a ground plane constraint.

View Processing

  • We have adapted the elegant background model but use a different metric for chromaticity distortion to better handle dark colors near the origin in RGB colorspace,
  • Raw frames are represented in the RGB colorspace model. Consider the expected value Ei=(Er.Eg.Eb)of a single pixel based on the current background model . The line passing through and the origin of RGB color space is the expected chrominance line.
  • The difference between the expected value and the actual measured frame pixel is decomposed into two parts.
  • Instead of the orthogonal distance we use the cosine of the angle between the expected chrominance line Ei and the line Mi formed from the measured point Mi and the originz

Combined Processing

  • Object tracking across cameras is used to interpretthe combined sets of time-stamped foreground blobs segmented from each video stream.
  • CLASSIFICATION :We are using two features to classify objects: area and compactness. The area feature is the number of pixels belonging to the object. and is defined as:
  • C= AREA /PERIMETER2
  • Determination of an abandoned package event requires a precise definition of what it means for a package to be abandoned.
  • Here we have a state machine diagram of the detecting abandoned packages

Abandoned package detection

Detecting Abandoned Packages in Multi-Camera Video Surveillance System

Alert Notification Snapshot

State machine for detecting packages

When an object appears that is classified as a package, it begins in the Start state.

The static state is entered when the velocity of the package becomes low enough. If the package does not have an owner, or if the distance between the owner , it enters the alone state.

When a thresholded amount of time has passed and the package object has remained stationary and isolated from its owner, we enter the Alert state, and an operator is notified.

When the owner returns, or if the package starts moving, we leave the Alert state and turn off the notification.

  • Object segmentation is the process of precisely determining which pixels belong to which objects in a singleframe of video.
  • Motion is used to distinguish objects from the background. Since each camera in a multi-camera environment is independent, object segmentation can be performed concurrently on each camera video stream.

Citation

Future work

Beynon, M.D.; Van Hook, D.J.; Seibert, M.; Peacock, A.; Dudgeon, D., "Detecting abandoned packages in a multi-camera video surveillance system," in Advanced Video and Signal Based Surveillance, 2003. Proceedings. IEEE Conference on , vol., no., pp.221-228, 21-22 July 2003

  • Future work involves performing view processing

in parallel on the capture host near each camera

before being sent to another host for combined

processing, which can dramatically improve the

processing frame rate, since most execution time

is presently spent in pixel level operations for

each camera.

Based on Jim Harvey's speech structures