【摘要】：隨之我國經(jīng)濟社會的不斷向前發(fā)展,城市中的人口以及汽車保有量不斷增多,城市交通問題日益突出。智能交通系統(tǒng)的中存儲著大量高清圖片,這些交通圖像數(shù)據(jù)中包含著大量有價值信息,是警察部門破案、刑偵以及解決交通問題的重要工具。但是,傳統(tǒng)的智能交通系統(tǒng)中,缺乏有效的方案來提取圖像數(shù)據(jù)中的信息,特別是對于交通場景下的行人、車輛等物體識別任務,目前仍是以人工識別方式為主,浪費人力警力,慢速低效。針對交通系統(tǒng)每天產(chǎn)生的海量圖片與智能交通系統(tǒng)物體識別能力不足之間的矛盾,我們設計實現(xiàn)了面向交通圖像數(shù)據(jù)的快速物體識別系統(tǒng),旨在利用先進的計算機技術,實現(xiàn)快速的交通圖像物體識別,提取有效信息,輔助警方工作。在調(diào)研過程中,發(fā)現(xiàn)卷積神經(jīng)網(wǎng)絡作為近年來計算機視覺領域的前沿技術,在物體定位、物體識別、物體分類等多個領域都取得了很好效果,受到了學術界與工業(yè)界的高度關注。因此,我們選擇利用卷積神經(jīng)網(wǎng)絡來進行物體識別任務,并且利用回歸思想結(jié)合卷積神經(jīng)網(wǎng)絡來減少中間操作,實現(xiàn)快速的物體識別。而且,考慮到GPU在圖像數(shù)據(jù)處理中的強大能力,在某些環(huán)節(jié)采用CUDA編程方式,進一步加快處理速度�；诰矸e神經(jīng)網(wǎng)絡的快速物體識別系統(tǒng),采用模塊化設計方案,各個模塊均選擇成熟技術進行開發(fā)。采用卷積神經(jīng)網(wǎng)絡作為物體識別技術,并利用GPU的高計算能力進一步加速識別過程,主要采用C/C++進行開發(fā),在卷積網(wǎng)絡模塊會引入CUDA編程來加快運行速度,實現(xiàn)快速高效識別。數(shù)據(jù)接入采用成熟的Inotify技術,采用主動方式去獲得相關數(shù)據(jù),進一步減少延遲。與其他相關系統(tǒng)的通信采用網(wǎng)絡通信方式,避免了因為系統(tǒng)異構(gòu)造成的通信障礙,而且能夠配合交通部門已經(jīng)建立的大數(shù)據(jù)存儲系統(tǒng)共同使用。本系統(tǒng)針對交通物體識別這一功能需求,利用回歸的思想設計了相關的卷積神經(jīng)網(wǎng)絡模型,填補了交通警察部門的智能交通系統(tǒng)在圖像相關業(yè)務方面的空缺,使得用戶可以更加快速高效地利用交通圖像數(shù)據(jù),進一步加強了警方在交通領域的管理能力,對于實現(xiàn)現(xiàn)代化的交通管理具有重要意義。
[Abstract]:Along with the development of our country's economy and society, the population in the city and the number of cars in the city are increasing, and the problem of urban traffic is becoming more and more serious. The intelligent transportation system stores a large number of high-definition pictures, which contain a large amount of valuable information, which is an important tool for the police department to break the case, criminal investigation and solve the traffic problem. However, in the traditional intelligent traffic system, there is a lack of effective scheme to extract the information in the image data, especially for pedestrian, vehicle and other objects in the traffic scene. Aiming at the contradiction between the mass picture produced by the traffic system and the insufficient recognition capability of the intelligent traffic system object every day, the rapid object recognition system aiming at the traffic image data is designed, and the aim of the invention is to realize the rapid traffic image object recognition by using the advanced computer technology, The effective information is extracted and the police are assisted to work. In the course of investigation, it is found that the convolution neural network has achieved good results in many fields such as object location, object recognition, object classification and so on, and has been highly concerned by the academic and industry. Therefore, we choose to use the convolution neural network to carry on the object recognition task, and use the regression idea to combine the convolution neural network to reduce the intermediate operation and realize the fast object recognition. In addition, considering the powerful ability of the GPU in the image data processing, the CUDA programming mode is adopted in some links, and the processing speed is further accelerated. The rapid object recognition system based on the convolution neural network adopts the modular design scheme, and each module selects the mature technology for development. In this paper, the convolution neural network is used as the object recognition technology, and the high computing power of the GPU is used to further accelerate the identification process, and the C/ C ++ is mainly used for development, and the convolution network module can introduce the CUDA programming to speed up the running speed and realize the fast and high-efficiency identification. Data access adopts the mature Intify technology, and the active mode is adopted to obtain the relevant data, and the delay is further reduced. The communication with other relevant systems adopts the network communication mode, so that the communication barrier caused by the heterogeneous system is avoided, and the large-data storage system which has been established by the traffic department can be used for common use. in that system, the function requirement of the traffic object is identify, the relevant convolution neural network model is designed by using the thought of the regression, and the vacancy of the intelligent traffic system in the traffic police department in the image-related business is filled, So that the user can use the traffic image data more quickly and efficiently, and further strengthens the management capability of the police in the traffic field, and is of great significance for realizing the modern traffic management.
【學位授予單位】：山東大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP311.52;TP183

【參考文獻】

相關期刊論文 前1條

1 曹林林;李海濤;韓顏順;余凡;顧海燕;;卷積神經(jīng)網(wǎng)絡在高分遙感影像分類中的應用[J];測繪科學;2016年09期

相關碩士學位論文 前4條

1 汪濟民;基于卷積神經(jīng)網(wǎng)絡的人臉檢測和性別識別研究[D];南京理工大學;2015年

2 趙明;智能交通系統(tǒng)數(shù)據(jù)挖掘與應用[D];山東大學;2014年

3 劉偉;電子發(fā)票校驗與呈遞系統(tǒng)的設計與實現(xiàn)[D];山東大學;2013年

4 唐亞光;Linux平臺自動軟件發(fā)布系統(tǒng)的設計和實現(xiàn)[D];浙江大學;2009年

，

本文編號：2492678