Title :
New performance measures in object detection
Author :
Kuzmin, Sergey A.
Author_Institution :
Dept. of Radio Tech. & Optoelectron. Complexes, St. Petersburg State Univ. of Aerosp. Instrum., St. Petersburg, Russia
Abstract :
The article is devoted to new approaches in accuracy evaluation of information retrieval systems (with tests made for motion detection algorithms). First part of article is devoted to studying of how good is tested motion detection algorithm. New elements are added to ROC for computing new measures. These elements are based on Equal Error Rate criterion. Second part of article is devoted to visualization of progress for multiple-step motion detection algorithms. Any algorithm of this type is a set of blocks and it is important to know which gain in accuracy was achieved in Nth step. Elements of design of multiple-step motion detection algorithms are discussed. Another discussed type of visualization is dedicated to boosting (combination of two and more algorithms).
Keywords :
error statistics; information retrieval systems; motion estimation; object detection; performance evaluation; sensitivity analysis; ROC; equal error rate criterion; information retrieval systems; multiple-step motion detection algorithms; object detection; performance measures; progress visualization; Algorithm design and analysis; Classification algorithms; Coordinate measuring machines; Estimation; Image edge detection; Motion detection; Visualization; Equal Error Rate; ROC; boosting; decision making; false acceptance rate; information retrieval; object detection; performance evaluation; true acceptance rate; visualization;
Conference_Titel :
Young Researchers in Electrical and Electronic Engineering Conference (EIConRusNW), 2015 IEEE NW Russia
Conference_Location :
St. Petersburg
Print_ISBN :
978-1-4799-7305-7
DOI :
10.1109/EIConRusNW.2015.7102240