mobile object detection through client-server based vote transfer
DESCRIPTION
Mobile Object Detection through Client-Server based Vote Transfer. CVPR 2012 poster. Outline. Introduction Frame detection Mobile application blue-print Experiment Conclusion. Introduction. Android OS. Introduction. Short video sequence. Introduction. Main Contribution: - PowerPoint PPT PresentationTRANSCRIPT
CVPR 2012 POSTER
Mobile Object Detection through Client-Server based
Vote Transfer
Outline
IntroductionFrame detectionMobile application blue-printExperimentConclusion
Introduction
Android OS
Introduction
Short video sequence
Introduction
Main Contribution: Novel hough forest based multi-frame object detection
framework
Vote transfer
Client-server framework
Frame detection
Single-Frame detection Hough forest [10]
[10] J. Gall and V. Lempitsky. Class-specific hough forests forobject detection. In CVPR, 2009.
Frame detection
P={L,c,d}
Frame detection
Multi-Frame detection Motivation Different express with single frame detection
Frame detection
Multi-Frame detection Vote transfer
Frame detection
Multi-Frame detection Vote transfer
Frame detection
Mobile application blue-print
Client-server
Experiment
Datasets A new multi-view dataset that we collected the Car Show Dataset introduced by Ozuysal et al
[19] http://www.eecs.umich .edu/vision/Mvproject.html
[19] Pose estimation for categoryspecific multiview object localization. In CVPR, 2009
Experiment
Vote transfer Giving each a weight Reference frame’s weight=1 Other frames’s weight= 2 -i/10 , i={10,20,30,40,50}
Experiment
Single vs Multi-frame Performance
Experiment
Single vs Multi-frame Performance
Experiment
Tracking analysis
Experiment
Image resolution
Experiment
Mobile platform: Client-Server analysisClient:
Motorola Atrix 4g dual-core phone Android 2.2
Image size:640*480Server:
2.4GHZ triple-core desktop
For more information to Motorola Atrix http://www.motorola.com/us/consumers/Motorola-ATRIX-4G/72112,en_US,pd.html?cgid=mobile-phones
Experiment
Mobile platform: Client-Server analysis Single frame
Multi frame
Conclusion
A new approach to multi-frame object detection using Hough Forest
Realistic implementation Client-server approach on mobile platform
About future work: Pose estimation, how view-point changes can foster pose estimation
Thanks for your listening.