【摘要】：交通情境識(shí)別又稱(chēng)為交通模式識(shí)別,是利用用戶(hù)的上下文信息對(duì)用戶(hù)所處的交通狀態(tài)的一種識(shí)別和感知,是人類(lèi)行為識(shí)別的子問(wèn)題。交通模式的自動(dòng)識(shí)別可以替代傳統(tǒng)的居民出行調(diào)查方式,更加便捷的獲取大量居民的出行方式信息數(shù)據(jù)。用于城市的交通規(guī)劃,緩解城市交通壓力以及提高人們的出行效率。本文探究使用深度學(xué)習(xí)的方法對(duì)手機(jī)傳感器進(jìn)行建模完成交通情境識(shí)別。本文首先研究了傳統(tǒng)的基于手機(jī)傳感器的交通模式識(shí)別方法,包括使用的手機(jī)傳感器類(lèi)型、數(shù)據(jù)流處理過(guò)程、以及傳統(tǒng)的分類(lèi)方法的性能。本文研究的交通模式分類(lèi)包括:公交、地鐵、出租、高鐵。根據(jù)研究需求采集并且構(gòu)建了相關(guān)交通模式識(shí)別的基準(zhǔn)數(shù)據(jù)集。共進(jìn)行255次采集包含15名采集人員,采集6種不同部位,總共7861分鐘的數(shù)據(jù)。在基準(zhǔn)測(cè)試數(shù)據(jù)集上,本文提出兩種交通模式識(shí)別方案:一、基于多層的RNN交通模式識(shí)別方案。方案對(duì)傳感器進(jìn)行預(yù)處理后提取簡(jiǎn)單的統(tǒng)計(jì)特征作為RNN網(wǎng)絡(luò)的輸入,使用多層或單層的lstm網(wǎng)絡(luò)提取時(shí)序特征用于交通模式識(shí)別,最終識(shí)別準(zhǔn)確率可以達(dá)到89%;二、結(jié)合CNN和RNN的交通模式識(shí)別方案,本方案通過(guò)將傳感器數(shù)據(jù)特征圖像化,生成activity image利用CNN自動(dòng)的提取特征并利用RNN網(wǎng)絡(luò)學(xué)習(xí)特征圖像的時(shí)序特征。最終識(shí)別準(zhǔn)確率可以到達(dá)78%。
[Abstract]:Traffic situation recognition, also known as traffic pattern recognition, is a kind of recognition and perception of the user's traffic state using the user's context information. It is a sub-problem of human behavior recognition. The automatic identification of traffic patterns can replace the traditional residents' travel survey and obtain a large number of residents' travel mode information data more conveniently. It is used in urban traffic planning, relieving traffic pressure and improving people's travel efficiency. This paper explores the use of depth learning to model mobile phone sensors to complete traffic situation recognition. Firstly, this paper studies the traditional traffic pattern recognition methods based on mobile phone sensors, including the types of mobile phone sensors used, the process of data flow processing, and the performance of the traditional classification methods. This paper studies the classification of traffic patterns including: public transport, subway, taxi, high-speed rail. According to the needs of the research, the benchmark data set of traffic pattern recognition is constructed. A total of 7861 minutes of data were collected from 6 different parts. Based on the benchmark data set, this paper proposes two traffic pattern recognition schemes: first, RNN traffic pattern recognition scheme based on multi-layer. After preprocessing the sensor, the scheme extracts simple statistical features as input of RNN network, and uses multi-layer or single-layer lstm network to extract time series features for traffic pattern recognition. Finally, the recognition accuracy can reach 89 parts. Combined with the traffic pattern recognition scheme of CNN and RNN, this scheme uses the sensor data feature image to generate activity image automatically extract features using CNN and use RNN network to learn the temporal features of feature images. The final recognition accuracy can reach 78.
【學(xué)位授予單位】：浙江大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類(lèi)號(hào)】：U495;TP391.4

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 遲鐵軍;高鵬;;國(guó)外智能交通系統(tǒng)發(fā)展?fàn)顩r分析及對(duì)我國(guó)的啟示[J];黑龍江交通科技;2009年02期

2 蔣慧強(qiáng);李資;;模式識(shí)別技術(shù)及其在消防通信中的應(yīng)用[J];科技信息(學(xué)術(shù)研究);2007年35期

，

本文編號(hào)：2199513