Combined shape and feature-based video analysis and its application to non-rigid object tracking

Combined shape and feature-based video analysis and its application to non-

rigid object tracking

資訊碩一 10077034 蔡勇儀2011/11/01 @LAB603

Introduction Method

Background generation and updating Detection of moving object Shape control points Combined shape and feature-based object

tracking Object occlusion

Result Conclusions

Outline

Object motion detect is an important issue of computer vision.

Many challenges Complex background More object motion Occlusion Illumination change Dynamic shading Camera jitter …

Introduction – object motion

Active shape model(ASM) Pre-model object’s shape Priori trained shape information Manually determined landmark point Can’t real time

Non-prior training active feature model(NPT-AFM) Consider feature point without object shape Improve computational efficiency Doesn’t utilise background information

Introduction – methods(1/2)

Block matching algorithm(BMA) Block matching between two frame Direct matching nature simplifies motion Preserves object’s feature which can’t

be easily parameterized

Poor performance with non-rigid shapes and similar patterns to the background.

Introduction – methods(2/2)

1. Background generation2. Motion detection and SCP extraction3. Object shape tracking modules

Introduction – This paper method

Introduction Method

Background generation and updating Detection of moving object Shape control points Combined shape and feature-based

object tracking Result Conclusions

Outline

Use median filter & BMA Define sum of absolute

difference(SAD) and threshold(0.05)

Find background(Static)

Method – Background generation

Method – Detection of moving object

Find feasible boundary

R represents the minimum rectangular box enclosing the object.

Method – Shape control points(1/2)

Build SCP set

K: interval of skipping redundant SCPs

Method – Shape control points(2/2)

Get block SCP

If object deformation, occlusion(25%)… CBMA – computing distances among

SCPs PBMA – fix motion region

Method - Combined shape and feature-based object tracking

Method - Summary

Introduction Method

Background generation and updating Detection of moving object Shape control points Combined shape and feature-based

object tracking Result Conclusions

Outline

Result(1/2)

Result(2/2)

Introduction Method

Background generation and updating Detection of moving object Shape control points Combined shape and feature-based

object tracking Result Conclusions

Outline

BMA & CBMA

The number of SCPs

Optimal region(feature histogram)

Conclusions

Source ：IET Image Process, 2011, Vol.5, Iss.1, pp.87-100

Thanks for your listening