基于視頻的交通沖突自動判別技術(shù)研究
發(fā)布時間：2018-12-27 13:48
【摘要】：伴隨著中國進入汽車時代，交通事故也增多起來，由此帶來了巨大的經(jīng)濟損失。因此，有效的對交叉口或路段進行交通安全性評價成為一個重要的研究課題。由于交通事故的后發(fā)性，用交通事故評價交通安全，在實時性和準確性方面稍顯不足。因此，基于交通沖突的交通安全評價方法得到了廣泛的研究。 雖然基于交通沖突的方法能夠較好的評價交通安全性，但是目前的交通沖突的采集方法主要是通過人工觀察的手段，這種方法實時性差且耗時費力，由此帶來一些測量上的不便。因此本文提出了基于視頻的交通沖突自動判別技術(shù)，該技術(shù)能夠自動判別交通沖突的發(fā)生，，克服了基于人工觀測方法的諸多弊端。 目前基于視頻的交通沖突自動判別系統(tǒng)的研究存在著諸多問題，如目標檢測不準確，跟蹤效果不好，沖突判別正確率不高等。因此本文針對目前交通沖突自動判別系統(tǒng)研究中存在的問題，做出了深入的研究。主要工作如下： （1）基于背景差分的目標檢測算法 目標檢測是進行交通沖突自動判別的第一步，目標檢測的準確程度直接決定著沖突判別的準確程度。本文對比了現(xiàn)有各個目標檢測算法的優(yōu)劣，最后決定采用背景差分算法提取目標。該算法首先利用背景初始化算法提取背景，并且利用背景更新模型更新背景。然后利用背景差分算法得到二值化前景圖像，并用連通區(qū)域標定算法得到各個前景目標。最后，利用目標分類算法得到目標的分類，實現(xiàn)目標檢測。通過實驗分析，本算法取得了良好的檢測效果。 （2）基于在線學(xué)習(xí)的目標跟蹤算法 接著要進行目標跟蹤，本文提出了改進的在線增強跟蹤算法。由于原始的在線增強跟蹤算法存在著實時性差和對左轉(zhuǎn)目標漂移的問題，因此本文對其進行了改進。首先，本文提出了一個級聯(lián)分類器提高跟蹤速度，然后提出了一個主方向模型改進跟蹤效果，解決了跟蹤中存在的漂移問題，最后提出了目標位置預(yù)測模型減少搜索區(qū)域，進一步提高跟蹤的實時性。通過實驗分析，本文改進的在線增強跟蹤算法較傳統(tǒng)在線增強算法跟蹤速度快，跟蹤精度高。 （3）交通沖突自動判別算法 在獲取了目標檢測區(qū)域內(nèi)的所有目標的軌跡序列和速度序列數(shù)據(jù)后，就要進行實時自動的交通判別。首先本文建立了基于臨界距離的交通沖突判別模型，能夠?qū)煌_突進行判別。然后基于視頻處理技術(shù)，建立了交通沖突的判別流程。將傳統(tǒng)檢測方法和本文基于視頻的自動判別方法進行對比實驗，結(jié)果表明：本文所提出的方法判別速度更快，準確度更高。 總而言之，本文進一步深化了對自動交通沖突判別領(lǐng)域的研究，所提出的算法與現(xiàn)有方法相比更具實時性和準確性，能夠為交叉口或路段安全性評價服務(wù)，具有重要的理論與實用價值。
[Abstract]:With China entering the automobile age, traffic accidents have also increased, which has brought huge economic losses. Therefore, it is an important research topic to evaluate the traffic safety of intersection or section effectively. Due to the lateness of traffic accidents, the evaluation of traffic safety by traffic accidents is a little insufficient in real time and accuracy. Therefore, traffic safety evaluation method based on traffic conflict has been widely studied. Although the method based on traffic conflict can better evaluate the traffic safety, the current traffic conflict collection method is mainly through the means of manual observation, this method is poor real-time and time-consuming. This brings some inconvenience in measurement. Therefore, this paper proposes a video based automatic discrimination technique for traffic conflicts, which can automatically distinguish the occurrence of traffic conflicts and overcome many disadvantages of artificial observation methods. At present, there are many problems in the research of traffic conflict automatic discriminant system based on video, such as inaccurate target detection, poor tracking effect, low accuracy of conflict discrimination and so on. Therefore, this paper makes an in-depth study on the existing problems in the research of traffic conflict automatic discrimination system. The main work is as follows: (1) Target detection algorithm based on background difference is the first step of automatic traffic conflict discrimination. The accuracy of target detection directly determines the accuracy of conflict discrimination. This paper compares the advantages and disadvantages of the existing target detection algorithms, and finally decides to use background differential algorithm to extract the target. Firstly, the background initialization algorithm is used to extract the background, and the background update model is used to update the background. Then the background difference algorithm is used to obtain the binary foreground image and the connected region calibration algorithm is used to obtain each foreground target. Finally, the target classification algorithm is used to achieve target detection. Through experimental analysis, the algorithm has achieved a good detection effect. (2) the target tracking algorithm based on online learning is proposed in this paper. Since the original on-line enhanced tracking algorithm has the problems of poor real-time performance and drift to the left turn target, this paper improves the algorithm. Firstly, a cascade classifier is proposed to improve the tracking speed, then a main direction model is proposed to improve the tracking effect, which solves the drift problem in tracking. Finally, the target location prediction model is proposed to reduce the search area. Further improve the real-time tracking. Through the experimental analysis, the improved on-line enhancement tracking algorithm is faster than the traditional on-line enhancement algorithm, and the tracking accuracy is high. (3) the traffic conflict automatic discriminant algorithm will carry on the real-time automatic traffic discrimination after obtaining the track sequence and velocity sequence data of all the targets in the target detection area. Firstly, a traffic conflict discriminant model based on critical distance is established, which can distinguish traffic conflict. Then, based on the video processing technology, the traffic conflict identification process is established. By comparing the traditional detection method with the automatic discriminant method based on video in this paper, the results show that the method proposed in this paper is faster and more accurate. In a word, this paper further deepens the research on the field of automatic traffic conflict discrimination. Compared with the existing methods, the proposed algorithm is more real-time and accurate, and can serve for the safety evaluation of intersections or sections. It has important theoretical and practical value.
【學(xué)位授予單位】：吉林大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：U491.265;U495
本文編號：2393164
[Abstract]:With China entering the automobile age, traffic accidents have also increased, which has brought huge economic losses. Therefore, it is an important research topic to evaluate the traffic safety of intersection or section effectively. Due to the lateness of traffic accidents, the evaluation of traffic safety by traffic accidents is a little insufficient in real time and accuracy. Therefore, traffic safety evaluation method based on traffic conflict has been widely studied. Although the method based on traffic conflict can better evaluate the traffic safety, the current traffic conflict collection method is mainly through the means of manual observation, this method is poor real-time and time-consuming. This brings some inconvenience in measurement. Therefore, this paper proposes a video based automatic discrimination technique for traffic conflicts, which can automatically distinguish the occurrence of traffic conflicts and overcome many disadvantages of artificial observation methods. At present, there are many problems in the research of traffic conflict automatic discriminant system based on video, such as inaccurate target detection, poor tracking effect, low accuracy of conflict discrimination and so on. Therefore, this paper makes an in-depth study on the existing problems in the research of traffic conflict automatic discrimination system. The main work is as follows: (1) Target detection algorithm based on background difference is the first step of automatic traffic conflict discrimination. The accuracy of target detection directly determines the accuracy of conflict discrimination. This paper compares the advantages and disadvantages of the existing target detection algorithms, and finally decides to use background differential algorithm to extract the target. Firstly, the background initialization algorithm is used to extract the background, and the background update model is used to update the background. Then the background difference algorithm is used to obtain the binary foreground image and the connected region calibration algorithm is used to obtain each foreground target. Finally, the target classification algorithm is used to achieve target detection. Through experimental analysis, the algorithm has achieved a good detection effect. (2) the target tracking algorithm based on online learning is proposed in this paper. Since the original on-line enhanced tracking algorithm has the problems of poor real-time performance and drift to the left turn target, this paper improves the algorithm. Firstly, a cascade classifier is proposed to improve the tracking speed, then a main direction model is proposed to improve the tracking effect, which solves the drift problem in tracking. Finally, the target location prediction model is proposed to reduce the search area. Further improve the real-time tracking. Through the experimental analysis, the improved on-line enhancement tracking algorithm is faster than the traditional on-line enhancement algorithm, and the tracking accuracy is high. (3) the traffic conflict automatic discriminant algorithm will carry on the real-time automatic traffic discrimination after obtaining the track sequence and velocity sequence data of all the targets in the target detection area. Firstly, a traffic conflict discriminant model based on critical distance is established, which can distinguish traffic conflict. Then, based on the video processing technology, the traffic conflict identification process is established. By comparing the traditional detection method with the automatic discriminant method based on video in this paper, the results show that the method proposed in this paper is faster and more accurate. In a word, this paper further deepens the research on the field of automatic traffic conflict discrimination. Compared with the existing methods, the proposed algorithm is more real-time and accurate, and can serve for the safety evaluation of intersections or sections. It has important theoretical and practical value.
【學(xué)位授予單位】：吉林大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：U491.265;U495
【參考文獻】
相關(guān)期刊論文 前10條
1 藺志青,郭軍;貝葉斯分類器在手寫漢字識別中的應(yīng)用[J];電子學(xué)報;2002年12期
2 王臺生;;基于HMM的背景建模初始化算法研究[J];硅谷;2011年19期
3 羅石貴,周偉;路段交通沖突技術(shù)研究[J];公路交通科技;2001年01期
4 顧國慶,許彥冰;數(shù)字圖像區(qū)域標定的方法[J];上海理工大學(xué)學(xué)報;2001年04期
5 李志慧;張長海;曲昭偉;魏巍;王殿海;;交通流視頻檢測中背景初始化算法[J];吉林大學(xué)學(xué)報(工學(xué)版);2008年01期
6 萬纓;韓毅;盧漢清;;運動目標檢測算法的探討[J];計算機仿真;2006年10期
7 盧官明;謝雙;;自適應(yīng)背景更新及運動目標檢測算法[J];南京郵電大學(xué)學(xué)報(自然科學(xué)版);2013年03期
8 ;Traficon視頻檢測的世界級參考標準——TrafiCam一體化路口視頻車輛檢測器[J];中國交通信息產(chǎn)業(yè);2007年02期
9 羅石貴,周偉;路段交通沖突的調(diào)查技術(shù)[J];長安大學(xué)學(xué)報(自然科學(xué)版);2003年01期
10 李強;趙亦工;高永晶;;基于背景差分的動目標檢測算法研究[J];遙測遙控;2007年06期
相關(guān)博士學(xué)位論文 前4條
1 曲昭偉;混合交通視頻檢測算法研究[D];吉林大學(xué);2009年
2 胡宏宇;基于視頻處理的交通事件識別方法研究[D];吉林大學(xué);2010年
3 曾勇;廣義近鄰模式分類研究[D];上海交通大學(xué);2009年
4 江晟;混合交通視頻檢測關(guān)鍵技術(shù)研究[D];吉林大學(xué);2013年
本文編號：2393164
本文鏈接：http://sikaile.net/kejilunwen/jiaotonggongchenglunwen/2393164.html
教材專著
