本文選題：車型識(shí)別 + 車輛測(cè)量�。� 參考：《長(zhǎng)安大學(xué)》2017年碩士論文

【摘要】：車型的自動(dòng)識(shí)別技術(shù)是ITS系統(tǒng)的關(guān)鍵技術(shù)之一,而車輛外形的三維尺寸能夠?yàn)檐囆妥R(shí)別提供可靠有效的數(shù)據(jù)基礎(chǔ)。傳統(tǒng)的車輛外形測(cè)量方法主要依賴于激光雷達(dá),它的測(cè)量精度較高,但容易受天氣和大氣的影響,而且成本較高,不利于大規(guī)模的實(shí)際應(yīng)用。本文結(jié)合交通應(yīng)用場(chǎng)景,研究了一種基于圖像處理的車輛外形測(cè)量方法,通過圖像處理技術(shù),實(shí)現(xiàn)車輛外形的長(zhǎng)寬高測(cè)量。其中,針對(duì)不同的應(yīng)用場(chǎng)景,本文提出了基于平面圖像和基于深度圖像的兩種實(shí)現(xiàn)方案,它們都采用相機(jī)近距離安裝的方式獲取清晰的車輛圖像,然后利用一些先驗(yàn)知識(shí)和圖像處理方法獲得車輛外表面的三維坐標(biāo)。再根據(jù)車輛運(yùn)動(dòng)過程中的序列圖像,通過圖像配準(zhǔn)的方法拼接出完整的車輛圖像,最后實(shí)現(xiàn)車輛外形的三維測(cè)量。本文的研究?jī)?nèi)容主要有:1.本文結(jié)合交通應(yīng)用場(chǎng)景,研究了能夠有效利用現(xiàn)場(chǎng)信息的基于消失點(diǎn)的標(biāo)定算法,實(shí)現(xiàn)了相機(jī)的非現(xiàn)場(chǎng)標(biāo)定,并通過實(shí)驗(yàn)檢驗(yàn)了這種標(biāo)定方法的準(zhǔn)確性。2.本文研究了基于圖像獲取車輛三維坐標(biāo)的方法。在基于平面圖像的方案中,本文利用車輛到相機(jī)的水平距離,及中大型車輛的側(cè)面近似垂直于地面的特點(diǎn),通過逆投影的方法獲取車輛側(cè)面的三維坐標(biāo)。在基于深度圖像的方案中,本文通過深度信息和相機(jī)成像的幾何關(guān)系,獲取車輛外表面的三維點(diǎn)云,并通過實(shí)驗(yàn)分析了所獲取三維坐標(biāo)的誤差及準(zhǔn)確性。3.對(duì)于數(shù)米、甚至十幾米的中大型車輛,相機(jī)無(wú)法通過一幀圖像得到完整車輛圖像的問題,本文根據(jù)車輛在運(yùn)動(dòng)過程中的序列圖像,研究了車輛圖像的配準(zhǔn)方法,并提出了一種基于剛體運(yùn)動(dòng)一致性的約束方法,有效篩選剔除了錯(cuò)誤匹配點(diǎn),改善了車輛圖像配準(zhǔn)的結(jié)果,并實(shí)現(xiàn)了車輛圖像的完整拼接。最后,針對(duì)完整的帶有三維坐標(biāo)信息的車輛圖像,本文介紹了車輛外形測(cè)量的具體方法,并通過實(shí)驗(yàn)檢驗(yàn)了該方法的測(cè)量結(jié)果,分析了測(cè)量誤差的主要因素。
[Abstract]:Automatic vehicle recognition is one of the key technologies in its system, and the 3D dimension of vehicle shape can provide a reliable and effective data basis for vehicle recognition. The traditional vehicle shape measurement method mainly depends on the lidar. It has high accuracy, but it is easy to be affected by the weather and atmosphere, and the cost is high, which is not conducive to large-scale practical application. In this paper, a method of vehicle shape measurement based on image processing is studied based on the traffic application scene. The length, width and height of vehicle shape are measured by image processing technology. Among them, for different application scenarios, this paper proposes two implementation schemes based on plane image and depth image, both of which obtain clear vehicle images by using the method of camera close installation. Then some prior knowledge and image processing method are used to obtain the three-dimensional coordinates of the outer surface of the vehicle. According to the sequence images of vehicle motion, the complete vehicle images are stitched up by image registration method, and finally the 3D measurement of vehicle shape is realized. The main contents of this paper are as follows: 1. In this paper, the vanishing point based calibration algorithm which can effectively utilize the field information is studied, and the off-site calibration of the camera is realized. The accuracy of this calibration method is verified by experiments. In this paper, the method of obtaining three-dimensional coordinate of vehicle based on image is studied. In the scheme based on plane image, using the horizontal distance from vehicle to camera and the characteristic that the side of medium and large vehicle is approximately perpendicular to the ground, the 3D coordinate of vehicle side is obtained by inverse projection. In the scheme based on depth image, the 3D point cloud on the external surface of the vehicle is obtained by the geometric relation between the depth information and the camera imaging, and the error and accuracy of the obtained 3D coordinate are analyzed by experiments. For medium and large vehicles with several meters or even more than ten meters, the camera can not get a complete vehicle image through a frame image. In this paper, the registration method of vehicle image is studied according to the sequence image of the vehicle in the process of motion. A constraint method based on rigid body motion consistency is proposed, which can effectively filter out the wrong matching points, improve the result of vehicle image registration, and realize the complete mosaic of vehicle images. Finally, aiming at the complete vehicle image with 3D coordinate information, this paper introduces the specific method of vehicle shape measurement, and tests the results of the method through experiments, and analyzes the main factors of measurement error.
【學(xué)位授予單位】：長(zhǎng)安大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：U495;TP391.41

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 吳彤;劉嘉新;;基于車輛側(cè)向特征的視頻監(jiān)控車型分類的研究[J];儀表技術(shù);2016年02期

2 張紅兵;李海林;黃曉婷;馬守磊;;基于車前臉HOG特征的車型識(shí)別方法研究與實(shí)現(xiàn)[J];計(jì)算機(jī)仿真;2015年12期

3 華莉琴;許維;王拓;馬瑞芳;胥博;;采用改進(jìn)的尺度不變特征轉(zhuǎn)換及多視角模型對(duì)車型識(shí)別[J];西安交通大學(xué)學(xué)報(bào);2013年04期

4 陳浩;吳慶祥;王穎;林梅燕;蔡榮太;;基于脈沖神經(jīng)網(wǎng)絡(luò)模型的車輛車型識(shí)別[J];計(jì)算機(jī)系統(tǒng)應(yīng)用;2011年04期

5 馬蓓;張樂;;基于紋理特征的汽車車型識(shí)別[J];電子科技;2010年02期

6 秦鐘;;基于圖像不變矩特征和BP神經(jīng)網(wǎng)絡(luò)的車型分類[J];華南理工大學(xué)學(xué)報(bào)(自然科學(xué)版);2009年02期

，

本文編號(hào)：2072951