基于DSP嵌入式平台多路实时视频拼接技术

发布时间：2018-06-20 20:39

本文选题：全景环视系统 + 嵌入式系统　；参考：《上海交通大学》2013年硕士论文

【摘要】：实时图像拼接技术一直以来都是计算机视觉领域的研究热点之一。近年来随着汽车电子的发展，出现了车载环视技术，可以将汽车周围的环境以一幅鸟瞰图显示给驾驶者，提高行车安全。实时图像拼接技术是实现车载环视系统的关键。然而，目前有关图像拼接的研究大多以PC为实现平台，鲜见其在车载嵌入式系统，比如DSP处理器系统上的实现。本文针对图像拼接进行了算法研究，提出了车载实时图像拼接的系统架构，，选用SURF算法进行图像匹配，用RANSAC算法进行图像拼接并利用加权平均法进行图像融合。在核心算法层面，为了改善图像拼接算法的时间效率，本文在分析了SURF算法原理的基础上，提出了从方框滤波器权值，特征点周围子区域的选择以及子区域内Haar小波变换的采样点范围三方面进行算法的参数配置优化，并采用控制变量法分析了在不同的SURF参数配置下，对算法特征点匹配的时效性、准确性和鲁棒性等性能的影响，综合通过仿真实验结果，给出了SURF算法参数的选择策略。在系统算法结构层面，通过对实验结果的详细分析，本文对算法进行了逐步改进，采用改进拼接流程和增加配准图像的方法减小了图像定位的误差，并针对环形鸟瞰拼接的独特需求，设计了合适的拼接顺序和融合权值计算方式，改善了图像的最终显示效果，同时兼顾了算法的时间效率。改进后的算法具有较高的准确度，融合效果好，能显示一张完整无缝的环形图像，并满足系统的实时性需要。在系统实现方面，本文研究了基于TI公司的DaVinci处理器实现一个视频图像拼接系统，并提出了一个基于DSP系统的软件系统架构，在该架构下DSP端程序主要负责图像拼接的核心算法，ARM端程序主要负责通过V4L2驱动控制IO设备，两者通过访问共享内存实现图像数据的高效传递，并尽可能减少了互斥所引起的程序挂起时间。最后，本文说明了拼接算法在DSP上的实现方案及注意事项，并给出了测试结果。相比之前的研究，本课题实现了基于DSP嵌入式系统的全景图像拼接系统，并针对系统实时性的要求对算法进行了相应的改进。本文提出的SURF算法参数选择策略，可以根据不同图像拼接需求调整算法，例如在针对实时视频流的拼接上，采用合理的参数可以在保证SURF算法拼接精度的基础上，拼接效率提高30%左右。参数配置理论使得SURF算法能够更广泛的应用于不同的图像拼接领域。
[Abstract]:Real-time image mosaic technology has always been one of the hotspots in the field of computer vision. In recent years, with the development of automobile electronics, the technology of vehicle-mounted surround view has emerged, which can display the surrounding environment of the vehicle to the driver with an aerial view to improve the safety of driving. Real-time image stitching is the key to the realization of vehicle-mounted circle view system. However, at present, most of the researches on image mosaic are based on PC, and few of them are implemented on on-board embedded system, such as DSP processor system. In this paper, the algorithm of image mosaic is studied, and the system architecture of vehicle real time image mosaic is proposed. The SURF algorithm is used for image matching, the RANSAC algorithm is used for image stitching and the weighted average method is used for image fusion. At the core algorithm level, in order to improve the time efficiency of image stitching algorithm, based on the analysis of the principle of surf algorithm, the weight value of the filter is proposed. The selection of sub-region around the feature point and the sampling point range of Haar wavelet transform in the sub-region are optimized. The control variable method is used to analyze the different SURF parameter configurations. By synthesizing the simulation results, the selection strategy of the parameters of the surf algorithm is given, which affects the timeliness, accuracy and robustness of the feature point matching of the algorithm. At the structure level of the system algorithm, the algorithm is improved step by analyzing the experimental results in detail, and the error of image location is reduced by improving the stitching process and increasing the registration of images. According to the unique demand of annular bird 's-eye view stitching, a suitable order of stitching and the calculation of fusion weights are designed to improve the final display effect of the image and to take into account the time efficiency of the algorithm. The improved algorithm has high accuracy and good fusion effect. It can display a complete and seamless annular image and meet the real-time needs of the system. In the aspect of system implementation, this paper studies the implementation of a video image mosaic system based on TI's DaVinci processor, and proposes a software architecture based on DSP system. In this framework, the DSP program is mainly responsible for image stitching, and arm program is mainly responsible for controlling IO device through V4L2 driver. Both of them can transfer image data efficiently by accessing shared memory. And minimize the mutex caused by the program hang-up time. Finally, this paper describes the implementation of the splicing algorithm on DSP and points for attention, and gives the test results. Compared with the previous research, this paper implements a panoramic image mosaic system based on DSP embedded system, and improves the algorithm to meet the real-time requirements of the system. The SURF algorithm parameter selection strategy proposed in this paper can adjust the algorithm according to different image mosaic requirements. For example, in the real-time video stream mosaic, reasonable parameters can be used to ensure the accuracy of SURF algorithm stitching. The efficiency of stitching is increased by about 30%. Parameter assignment theory makes the surf algorithm more widely used in different image stitching fields.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP368.1;TP391.41

【相似文献】