Integration of Sparse Reprentation and Symmetrical Feature for Vehicle Analysis 2015

瀏覽統計 Email 通知 RSS Feed

簡歷

基本資料

Project title

Code/計畫編號

MOST103-2221-E019-035-MY2

Translated Name/計畫中文名

結合稀疏表示法與對稱特徵之車輛分析系統

Project Coordinator/計畫主持人

Shyi-Chyi Cheng

Co-Investigator(s)/共同執行人

謝君偉

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=11274500

Year

2015

Start date/計畫起

05-02-0007

Expected Completion/計畫迄

31-07-2016

Bugetid/研究經費

694千元

ResearchField/研究領域

資訊科學--軟體

稀疏表示法(sparse representation)企圖找尋一組超完備(overcomplete)的基底來更有效率地表示樣本資料，廣泛用在許多視覺的應用上，此計畫將利用此表示法，應用到車輛偵測與分析上，特別是車型與顏色分析，因為在環境有陰影、遮蔽及擾動的背景等情況下，要偵測出車輛與對應的車型顏色，一直都是電腦視覺領域內的一個大難題，此計畫預計利用 SURF特徵點的對稱性，不須任何動量資訊，可從單影像中把車輛偵測出來，這方法不需背景重建，因此不受背景晃動的影響，且非常有效率。一旦車輛被偵測出來，接下來需找出有用的特徵來做車型與年份辨識，但車型有所謂多樣性(multiplicity) 與混淆(ambiguity)之問題，多樣性意味著同一個車型因為每年改款會有不同的形狀在路上，混淆性意味著不同車型常常有類似的形狀，為了解決上述兩個問題，我們提出格子切割的技巧，搭配 SURF特徵，來做車輛車型與年份辨識，在此二年計畫中，我們預計發展一個「運用對稱 SURF特徵之車輛辨識」系統，提出一個解決方案。此計畫可分成二個部分：(1)車輛偵測研究；(2) 車輛資訊辨識研究及系統整合。期望在這二年計畫中，藉由這些技術的研發，能為台灣發展出具有智慧與視覺能力的加值產品，進而提升台灣視訊安全產業的技術水準。此計畫所發展的監控技術可應用在許多方面，如安全監控、防盜及防恐等方面，並可置於車站、停車場及道路等公共區域，對進出之車輛作安全控管，以達到事件偵測警示與犯罪預防等目標。 Sparse representation uses an over-complete dictionary to represent objects and has been widely used and quite successful in many vision-based applications. This project plans to develip a new vehicle analysis system to recognize vehciles including their colors and types via this representation and symmetrical SURFs. To detection vehicles from roads, the proposed symmetrical descriptor is first applied to determine the ROI of each front vehicle from roads without using any motion features. Two advantages can be gained from this scheme, i.e., no need of background subtraction and the extreme efficiency for real-time applications. After that, two challenges in MMR should be addressed, i.e., the multiplicity and ambiguity problems. The multiplicity problem means one vehicle model often owns different model shapes on roads at the same time. The ambiguity problem means vehicles even made from different companies often share similar shapes. To treat the two problems, a grid division scheme is proposed to separate a vehicle to several grids and then different weak classifiers trained on these grids are integrated together to build a strong ensemble classifier. In this two-year project, we plan to develop a symmetric SURF-based vehicle analysis system. This project can be divided into two parts: (1) real-time vehicle detection; (2) vehicle information analysis. It is sincerely hoped all relevant developed techniques can explore more potential markets and offer more smart solutions for the enterprises in Taiwan and abroad as well.

Keyword(s)

稀疏表示法
車輛辨識
遮蔽
對稱 SURF
車型辨識
對稱轉換
車輛顏色辨識
sparse representation
vehicle recognition
occlusion
symmetric SURF
vehicle make and model recognition
symmetry transformation
vehicle color identification

DSpace CRIS

Integration of Sparse Reprentation and Symmetrical Feature for Vehicle Analysis 2015

基本資料

Description