Please use this identifier to cite or link to this item:
標題: Intelligent Real Time Traffic Mobility Monitoring and Analysis for Road Safety Applications
智慧行車安全監控嵌入式系統設計-子計畫四:應用於行車安全之智慧型即時路況監控與分析( I )
作者: 吳崇賓
關鍵字: 智慧行車;電子電機工程類;嵌入式系統設計;即時路況監控;車輛偵測;行人偵測;應用研究
In this sub-project, real time vehicle and pedestrian detection and tracking system for the intelligent Car Safety Monitoring will be proposed. The developed algorithm and its hardware architecture will consist of pre-processing and ROI selection, dynamic ROI, vehicle and pedestrian detection, and vehicle and pedestrian tracking. First, for the pre-processing and ROI selection, shake elimination, road detection, and reference point adjustment of ROI will be proposed. The interference of poor road conditions is expected to be effectively reduced. For the dynamic ROI, the project will focus on the research of the analysis of adaptive ROI selection. It is expected to apply to different traffic conditions. For the vehicle and pedestrian detection, depth-based target segmentation and selection for efficiently detecting vehicle and pedestrian will be presented. For the tracking of vehicle and pedestrian, low complexity window match method will be proposed to track the object accurately. The relative low complexity architecture based on the presented algorithm will be implemented and verified in the embedded development platform. Finally, cell-based VLSI design will be implemented and integrated in the embedded development platform which is proposed by the integrated project. In the end of project execution stage, the demo will be shown.

本子計畫主要針對智慧行車安全監控提出即時車輛與行人偵測與追蹤,包含四個關鍵技術有前處理與ROI區域選擇、動態ROI、車輛與行人偵測與車輛與行人追蹤等演算法開發及其硬體架構。首先,針對前處理與ROI區域選擇,本計畫將提出車輛晃動消除、道路偵測與ROI參考點調整,預期可有效降低道路不佳的干擾;針對動態ROI,本計畫將提出適應性ROI區域分析之研究,可適用於不同行車路況;針對車輛與行人偵測,本計畫將提出以深度為基礎之影像切割與目標物選取,預期可以有效偵測車輛與行人;針對車輛與行人追蹤,本計畫將提出低複雜度視窗匹配法,預期可以在較低計算量下有效的追蹤物件。本子計畫提出之演算法將發展相對應之低複雜度硬體架構與實現,並在嵌入式開發平台上驗證。最後將實現Cell-based VLSI設計並整合於總計劃規劃之嵌入式開發平台,做完整展示。
其他識別: NSC100-2220-E005-004
Appears in Collections:電機工程學系所

Show full item record

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.