Object detection using image reconstruction with PCA Source

  • Slides: 10
Download presentation
Object detection using image reconstruction with PCA Source: Image and Vision Computing. March 2007.

Object detection using image reconstruction with PCA Source: Image and Vision Computing. March 2007. Authors: Luis Malagon-Borja, Olac Fuentes. Reporter: Yen-Chang, Chen. Date: 2007/1/9. 1

Outline • Introduction • PCA classifier • Adding a Support Vector Machine classifier •

Outline • Introduction • PCA classifier • Adding a Support Vector Machine classifier • Experimental results • Conclusions 2

Introduction The detection syste Classifier based on Image Reconstruction with PCA Output≧ 0 =>

Introduction The detection syste Classifier based on Image Reconstruction with PCA Output≧ 0 => Pedestrian Output< 0 => Non-pedestrian SVM classifier Reduction of false detections by means of heuristics -Eliminating single detections -Eliminating nearby detections Input image Output image 3

PCA classifier Mean object of the set Sub-image u. Covariance matrix C The principal

PCA classifier Mean object of the set Sub-image u. Covariance matrix C The principal component. PC (The eigenvectors of C) p r The projection of the subimage u. The reconstruct ed image P The first k eigenvectors of PC. d Reconstruction error 4

PCA classifier Ex: The sets g e An image 5

PCA classifier Ex: The sets g e An image 5

Adding a Support Vector Machine classifier Support vector Hyperplane M Support vector 6

Adding a Support Vector Machine classifier Support vector Hyperplane M Support vector 6

Adding a Support Vector Machine classifier Input image Eliminating single detections Eliminating clustered detections

Adding a Support Vector Machine classifier Input image Eliminating single detections Eliminating clustered detections PCA classifier SVM classifier (1)Group the detections (2) Eliminating single detections (1)Define a region (2) Choice max Preference to keep 7

Experimental results The capability of the system for detecting people in still images with

Experimental results The capability of the system for detecting people in still images with cluttered backgrounds. 8

Experimental results ROC curves comparing the performance of out classifiers versus the best reported

Experimental results ROC curves comparing the performance of out classifiers versus the best reported in the literature. 9

Conclusions • Authors have presented an object detection system for static images. • This

Conclusions • Authors have presented an object detection system for static images. • This system is able to detect frontal and rear views of pedestrians, and usually it can also detect side views of pedestrians. 10