skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: BUILDING ROBUST APPEARANCE MODELS USING ON-LINE FEATURE SELECTION

Abstract

In many tracking applications, adapting the target appearance model over time can improve performance. This approach is most popular in high frame rate video applications where latent variables, related to the objects appearance (e.g., orientation and pose), vary slowly from one frame to the next. In these cases the appearance model and the tracking system are tightly integrated, and latent variables are often included as part of the tracking system's dynamic model. In this paper we describe our efforts to track cars in low frame rate data (1 frame/second) acquired from a highly unstable airborne platform. Due to the low frame rate, and poor image quality, the appearance of a particular vehicle varies greatly from one frame to the next. This leads us to a different problem: how can we build the best appearance model from all instances of a vehicle we have seen so far. The best appearance model should maximize the future performance of the tracking system, and maximize the chances of reacquiring the vehicle once it leaves the field of view. We propose an online feature selection approach to this problem and investigate the performance and computational trade-offs with a real-world dataset.

Authors:
 [1];  [1];  [1]
  1. Los Alamos National Laboratory
Publication Date:
Research Org.:
Los Alamos National Lab. (LANL), Los Alamos, NM (United States)
OSTI Identifier:
985888
Report Number(s):
LA-UR-07-0614
TRN: US201017%%66
DOE Contract Number:
AC52-06NA25396
Resource Type:
Conference
Resource Relation:
Conference: SPIE DEFENCE AND SECURITY 2007 SYMPOSIUM ; 200704 ; ORLANDO
Country of Publication:
United States
Language:
English
Subject:
99; AUTOMOBILES; ORIENTATION; PERFORMANCE; SECURITY; TARGETS

Citation Formats

PORTER, REID B., LOVELAND, ROHAN, and ROSTEN, ED. BUILDING ROBUST APPEARANCE MODELS USING ON-LINE FEATURE SELECTION. United States: N. p., 2007. Web.
PORTER, REID B., LOVELAND, ROHAN, & ROSTEN, ED. BUILDING ROBUST APPEARANCE MODELS USING ON-LINE FEATURE SELECTION. United States.
PORTER, REID B., LOVELAND, ROHAN, and ROSTEN, ED. Mon . "BUILDING ROBUST APPEARANCE MODELS USING ON-LINE FEATURE SELECTION". United States. doi:. https://www.osti.gov/servlets/purl/985888.
@article{osti_985888,
title = {BUILDING ROBUST APPEARANCE MODELS USING ON-LINE FEATURE SELECTION},
author = {PORTER, REID B. and LOVELAND, ROHAN and ROSTEN, ED},
abstractNote = {In many tracking applications, adapting the target appearance model over time can improve performance. This approach is most popular in high frame rate video applications where latent variables, related to the objects appearance (e.g., orientation and pose), vary slowly from one frame to the next. In these cases the appearance model and the tracking system are tightly integrated, and latent variables are often included as part of the tracking system's dynamic model. In this paper we describe our efforts to track cars in low frame rate data (1 frame/second) acquired from a highly unstable airborne platform. Due to the low frame rate, and poor image quality, the appearance of a particular vehicle varies greatly from one frame to the next. This leads us to a different problem: how can we build the best appearance model from all instances of a vehicle we have seen so far. The best appearance model should maximize the future performance of the tracking system, and maximize the chances of reacquiring the vehicle once it leaves the field of view. We propose an online feature selection approach to this problem and investigate the performance and computational trade-offs with a real-world dataset.},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Mon Jan 29 00:00:00 EST 2007},
month = {Mon Jan 29 00:00:00 EST 2007}
}

Conference:
Other availability
Please see Document Availability for additional information on obtaining the full-text document. Library patrons may search WorldCat to identify libraries that hold this conference proceeding.

Save / Share:
  • Vector selection, or instance selection as it is often called in the data mining literature, performs a critical task in the development of nonparametric, similarity based models. Nonparametric, similarity based modeling (SBM) is a form of 'lazy learning' which constructs a local model 'on the fly' by comparing a query vector to historical, training vectors. For large training sets the creation of local models may become cumbersome, since each training vector must be compared to the query vector. To alleviate this computational burden, varying forms of training vector sampling may be employed with the goal of selecting a subset ofmore » the training data such that the samples are representative of the underlying process. This paper describes one such SBM, namely auto-associative kernel regression (AAKR), and presents five traditional vector selection methods and one robust vector selection method that may be used to select prototype vectors from a larger data set in model training. The five traditional vector selection methods considered are min-max, vector ordering, combination min-max and vector ordering, fuzzy c-means clustering, and Adeli-Hung clustering. Each method is described in detail and compared using artificially generated data and data collected from the steam system of an operating nuclear power plant. (authors)« less
  • Managing a wide variety of mobile devices across multiple mobile operating systems is a security challenge for any organization [1, 2]. With the wide adoption of mobile devices to access work-related apps, there is an increase in third-party apps that might either misuse or improperly handle user s personal or sensitive data [3]. HTML5 has been receiving wide attention for developing cross-platform mobile apps. According to International Data Corporation (IDC), by 2015, 80% of all mobile apps will be based in part or wholly upon HTML5 [4]. Though HTML5 provides a rich set of features for building an app, itmore » is a challenge for organizations to deploy and manage HTML5 apps on wide variety of devices while keeping security policies intact. In this paper, we will describe an upcoming secure mobile environment for HTML5 apps, called Sencha Space that addresses these issues and discuss how it will be used to design and build a secure and cross-platform mobile mapping service app. We will also describe how HTML5 and a new set of related technologies such as Geolocation API, WebGL, Open Layers 3, and Local Storage, can be used to provide a high end and high performance experience for users of the mapping service app.« less
  • The Enterprise Housing Divestiture Project (EHDP) aims to identify cost-effective energy efficiency and conservation measures for Russian apartment buildings and to implement these measures in the entire stock of buildings undergoing divestiture in six cities. Short-term measurements of infiltration and exterior wall heat-loss coefficient were made in the cities of Cheropovets, Orenburg, Petrozavodsk, Ryazan, and Vladimir. Long-term monitoring equipment was installed in six or more buildings in the aforementioned and in the city of Volxhov. The results of these measurements will be used to calibrate models used to select optimal retrofit packages and to verify energy savings. The retrofit categoriesmore » representing the largest technical potential in these buildings are envelope, heat recovery, and heating/hot water system improvements. This paper describes efforts to establish a useful thermal model calibration process. The model structures and analytical methods for obtaining building parameters from time series weather, energy use, and thermal response data are developed. Our experience applying these methods to two, nominally identical 5-story apartment buildings in the city of Ryazan is presented. Building envelope UA?s inferred from measured whole-building thermal response data are compared with UA?s based on window and wall U-values (the latter obtained by ASTM in-situ measurements of 20 wall sections in various Ryazan panel buildings) as well. The UA's obtained by these completely independent measurements differ by less than 10%.« less
  • This paper proposes a regularized robust recursive least squares (R3LS) method for on-line estimation of power-system electromechanical modes based on synchronized phasor measurement unit (PMU) data. The proposed method utilizes an autoregressive moving average exogenous (ARMAX) model to account for typical measurement data, which includes low-level pseudo-random probing, ambient, and ringdown data. A robust objective function is utilized to reduce the negative influence from non-typical data, which include outliers and missing data. A dy-namic regularization method is introduced to help include a priori knowledge about the system and reduce the influence of under-determined problems. Based on a 17-machine simulation model,more » it is shown through the Monte-Carlo method that the proposed R3LS method can estimate and track electromechanical modes by effectively using combined typical and non-typical measurement data.« less