基于視頻的運(yùn)動車輛檢測與跟蹤

2.32萬字
自己原創(chuàng)的畢業(yè)論文，已經(jīng)通過校內(nèi)系統(tǒng)檢測，重復(fù)率低，僅在本站獨(dú)家出售，大家放心下載使用

摘 要
基于視頻的運(yùn)動車輛的檢測與跟蹤是智能交通信號系統(tǒng)和先進(jìn)交通事件管理系統(tǒng)的重要支撐，是智能交通系統(tǒng)中的一項(xiàng)重要課題。本文將圍繞視頻車輛檢測與跟蹤的相關(guān)技術(shù)展開研究，主要研究內(nèi)容及成果包括：
（1）交通場景前景提取算法的研究。分析了目前運(yùn)動前景檢測的幾種算法。針對一般的交通場景，提出了一種改進(jìn)的混合高斯模型來檢測場景中的運(yùn)動前景。所提算法能實(shí)時有效的進(jìn)行前景提取。
（2）運(yùn)動目標(biāo)分類算法的研究。結(jié)合目標(biāo)前景的形態(tài)特征與運(yùn)動特征，設(shè)計(jì)了基于多類支持向量機(jī)的分類算法，并用目標(biāo)軌跡軸上分類結(jié)果的概率統(tǒng)計(jì)進(jìn)行分類結(jié)果的修正。算法能夠?qū)⑶熬澳繕?biāo)劃分成小型車、大型車、行人及其自行車類型，并具備了較高的準(zhǔn)確率。
（3）運(yùn)動目標(biāo)跟蹤算法的研究。綜合考慮了目標(biāo)的位置、速度、形狀參數(shù)，將最短距離法和三次指數(shù)平滑預(yù)測算法相結(jié)合，進(jìn)行前后幀之間的參數(shù)匹配，完成跟蹤。

關(guān)鍵詞：背景建模 目標(biāo)分類 車輛跟蹤

Abstract
Technology of detection and tracking of vehicles based on video has become an important subject of intelligent transportation system. This paper launched a study on the key technology in traffic incident detection process. The main research contents and results include:
(1) Research on object detection algorithm for common traffic scene. This paper analyzes and studies some methods in background extraction and updating background. Focus on common traffic scenes, an improved Gaussian mixture model is proposed to detect moving targets in the dynamic scenes., which can extract the real-time background image and implement background real-time updating effectively.
(2) Research on traffic object classification and tracking algorithms. Combining morphological feature and moving feature of foreground, a classification algorithm based on multi-class support vector machine is designed. After that, statistical probability of classification results in target trajectory axis is used to modify the result. This algorithm can accurately divide foreground objects into small cars, large cars, pedestrians and bicycles. For target tracking problem, the position, velocity and shape parameter of target are all taken into account. Based on this, shortest distance method and three times of exponential smoothing prediction algorithm are combined to match parameters between the front frames and rear ones so that tracking can be accomplished.
(3) Research on moving vehicle shadows elimination. For target occlusion problem, this paper makes a study of gray-level distribution features of shadow and develops an effective shadow elimination method based on analyzing photometric gain properties of shadow and background region.

Keywords: background modeling; object classification; vehicle tracking

目 錄
摘 要 I
Abstract II
第一章 緒論 1
1.1研究背景 1
1.2研究綜述 2
1.2.1系統(tǒng)的發(fā)展歷程 2
1.2.2相關(guān)算法的研究現(xiàn)狀 4
1.3研究內(nèi)容及章節(jié)安排 5
1.4研究意義 7
1.5本章小結(jié) 7
第二章 運(yùn)動目標(biāo)檢測 8
2.1引言 8
2.2基于GMM的運(yùn)動目標(biāo)檢測 10
2.2.1背景估計(jì) 10
2.2.2 狀態(tài)估計(jì) 11
2.2.3 背景更新 14
2.3 試驗(yàn)結(jié)果及分析 15
第三章 運(yùn)動目標(biāo)分類 18
3.1 引言 18
3.2形態(tài)特征和運(yùn)動特征相結(jié)合的目標(biāo)特征表達(dá) 20
3.2.1 形態(tài)特征表達(dá) 21
3.2.2 運(yùn)動特征表達(dá) 26
3.3基于SVM的目標(biāo)分類 27
3.3.1 基本原理 27
3.3.2 分類器的構(gòu)建 30
3.3.3 分類器的訓(xùn)練 31
3.3.4 分類器的識別 32
3.4基于最大概率統(tǒng)計(jì)的類別修正 34
3.5結(jié)果分析 35
第四章 運(yùn)動車輛跟蹤 36
4.1引言 36
4.2狀態(tài)表達(dá) 37
4.3狀態(tài)估計(jì) 38
4.4特征匹配與更新 39
4.4試驗(yàn)結(jié)果及分析 39
第五章 總結(jié)及展望 44
參考文獻(xiàn) 45