非合作目標論文：基于DM6446的非合作目標位姿估計算法研究

1、非合作目標論文：基于DM6446的非合作目標位姿估計算法研究【中文摘要】隨著信息科學(xué)技術(shù)的發(fā)展,計算機視覺在越來越多的行業(yè)得到廣泛的應(yīng)用。計算機視覺按照目標種類可以分為合作目標視覺測量和非合作目標視覺測量,非合作目標視覺測量近幾年成為該領(lǐng)域研究熱點。本文主要研究工作著重于兩大方面:一方面是給出了兩種非合作目標位姿解算比較實用的算法,這兩種算法經(jīng)過推導(dǎo)和仿真實驗,驗證是可行的,而且效率要比現(xiàn)有方法高。另一方面是在設(shè)計的實驗平臺上實現(xiàn)了上面的一種位姿解算算法。針對單目非合作目標測量,給出了兩種算法。一個是改進的單目視覺相似迭代算法,改進的地方在最后的位姿計算上,前人通過奇異值分解的方法求解的,本文

2、是通過構(gòu)建新的坐標系來求解的,這種做法較簡單,編程也會容易很多。此算法的思路是將2D-3D之間的關(guān)系轉(zhuǎn)換為3D-3D之間的關(guān)系,然后通過求解景深和絕對方位求解位姿的。另一個是基于直線特征的單目視覺位姿解算,此算法也是在前人研究的基礎(chǔ)上做了相應(yīng)的改進,初始值給取是通過求解析解給出的,最后的位姿計算是用Newton迭代法求解的,經(jīng)仿真實驗驗證此算法正確無誤,而且結(jié)果精度在誤差允許的范圍內(nèi)。設(shè)計一套硬件實驗平臺,用來驗證算法的有效性和合理性。針對實際應(yīng)用的要求,首先在硬件的選擇上做了可行性分析,并考慮到以后軟件升級,在硬件選擇時還留了一定的余地。整個硬件系統(tǒng)包括目標體、CCD攝像機、DM6446處理

3、器、LCD液晶顯示屏、PC機,為方便檢測,又具有代表性,目標體設(shè)計為一長方體;考慮到圖像處理數(shù)據(jù)量較大,需高性能處理器,所以選擇了TI公司的DM6446處理器,這款芯片性能很高,是專門用來做圖像處理的。在實驗平臺上實現(xiàn)前面的算法。分別在圖像濾波、邊緣檢測、Hough變換、位姿計算上做了實驗,從反復(fù)的實驗中,總結(jié)出了圖像前期處理很重要,這部分的好壞直接影響后面位姿計算結(jié)果精度;還有就是編寫程序要考慮運行時間,盡量優(yōu)化程序,使時間盡可能的短。實驗結(jié)果表明:在DM6446實驗平臺上實現(xiàn)非合作目標位姿測量,達到了預(yù)期的效果,可以滿足實際的需求?！居⑽恼緼s the development of i

4、nformation technology,computer vision has been widely applied to numerous areas. In accordance with the objectives,computer vision can be divided into cooperative objectives vision measurement and Non-cooperative objectives vision measurement. In recent years, non-cooperative visual measurement beco

5、me one of the most popular research topics in this field. This thesis takes the background of 863:“Docking and separation technology of *”.This study focused on two main aspects:one is to give two practical algorithm of solving non-cooperative targets pose, after deduction and simulation experiment,

6、the two algorithm is feasible, and efficiency than algorithm in other documents ;the other hand, in the designed experimental platform, achieved a pose decoding algorithm .For the measurement of monocular and non-cooperative target pose,this paper gives two algorithms.one is to improved monocular vi

7、sion similar iteration algorithm, In the final pose computation, made improvements. Previous solved the problem using singular value decomposition method,this paper constructs a new coordinate to decode pose, this approach is simple, programming will also a whole lot easier. This algorithm converts

8、the relationship between 2D-3D to the relationship between 3D-3D,then gain the pose data by solving the depth of field and the absolute position value.Another one is to monocular vision pose decoding based on linear characteristics,this algorithm also makes the corresponding improvement on the basis

9、 of previous studies, and the initial value is given by calculating analytical solutions, , the final pose computation is verified by Newton iteration method .after the simulation experiments,the algorithm is correct,and the result is correct in allowed error.Designing a set of hardware experimental

10、 platform, used to verify the efficiency and rationality of the algorithm. In view of the actual application requirements, firstly, making the feasibility analysis in hardware choice,and considering the software upgrades,leaving some leeway when choose the hardware.The whole hardware systems include

11、 objective, CCD camera, processor DM6446, LCD screen, PC. For the conveni- ence of detection, the target is a cuboid; the amount of image processing data is larher, so needing a high-performance processor. finally choosing DM6446, the chip performa- nce is high,and it is dedicated to make image proc

12、essing.Realizing the algorithm in experimental platform. Doing experiments in image filtering,edge detecting,Hough transform, pose computation; from repeated experiments, knowing that Image pre-processing is very important,this part will impact on the result directly.considering the time of program

13、consumes, try to optimize program, make the time as short as possible.The results show that Non-cooperative objectives pose measurement have achieved expected result,and met the requirement of actual application.【關(guān)鍵詞】非合作目標 DM6446 單目視覺 邊緣檢測 位姿測量【英文關(guān)鍵詞】Non-cooperative DM6446 Mono-vision Edge Detection

14、 Position and attitude measure【目錄】基于DM6446的非合作目標位姿估計算法研究摘要4-5Abstract5-6第1章 緒論9-171.1 研究背景及意義9-101.2 國內(nèi)外相關(guān)領(lǐng)域的研究現(xiàn)狀10-131.2.1 視覺測量技術(shù)發(fā)展10-111.2.2 非合作目標測量應(yīng)用領(lǐng)域11-131.3 視覺測量實現(xiàn)平臺13-161.4 本文的主要研究內(nèi)容16-17第2章 單目非合作測量算法17-392.1 算法理論基礎(chǔ)17-202.1.1 圖像坐標系、攝像機坐標系與世界坐標系17-202.2 基于模型的單目相似迭代算法20-252.2.1 深度估計21-222.2.2 位姿解算22-252.3 基于SoftNeWton的單目視覺位姿測量25-382.3.1 模型建立26-272.3.2 迭代初始值27-292.3.3 Newton迭代法292.3.4 算法流程及仿真實驗29-382.4 本章小結(jié)38-39第3章 算法實現(xiàn)平臺39-503.1 硬件系統(tǒng)設(shè)計39-453.1.1 可行性分析39-403.1.2 系統(tǒng)設(shè)計方案40-453.2 系統(tǒng)軟件平臺45-493.2.1 Linux操作系統(tǒng)453.2.2 交叉