三维枪弹痕迹自动识别系统关键技术研究

	三维枪弹痕迹自动识别系统关键技术研究
	凌剑勇
学位类型	博士
导师	何昕
	2014-07
学位授予单位	中国科学院大学
学位专业	机械电子工程
摘要	枪弹痕迹比对是公安刑事侦查技术中的重要研究方向，利用枪弹痕迹检验学知识，通过识别子弹痕迹，来判断、查找发射枪支，能够为涉枪案件提供侦查过程中的重要物证。而目前国内主要的弹痕比对手段仍然采用基于比较显微镜人工识别方法，该方法效率很低，准确率不高，并不能满足弹痕识别的高效性、准确性的需要。随着计算机技术的发展，弹痕自动识别技术已经成为实现弹痕比对自动化的有效途径，弹痕识别技术就是利用子弹上枪弹痕迹的固有特征，通过数字图像处理、模式识别等技术鉴别子弹的发射枪支的技术。正式由于弹痕识别技术的重要性和实用性，已经受到国内外学者的重视。我国在弹痕比对领域的研究是近几年才发展起来的，相比国外发达国家的弹痕识别技术，我国的弹痕比对都是基于二维灰度图像基础上进行的研究，由于二维灰度图像在数据采集过程中受光照强度、角度等外界环境因素影响较大，且二维图像并不能反映弹痕的形貌信息，所以在实际研究中，其识别结果的可靠性无法保证，最终两发子弹的判别工作还需要通过人工进行，识别效率并没有实质性的提高。而通过对国外相关领域的研究，本文发现子弹表面的深度纹理信息是弹痕识别的重要研究对象，所以本文根据弹痕比对技术的要求，即在保证高精度测量同时，保证避免对子弹纹理痕迹的二次破坏。采用了一套光学非接触三维测量系统对弹痕数据进行采集，能同时获取高精度的弹痕三维形貌数据和二维灰度图像，并主要介绍了该设备的原理、结构、标定过程及弹痕ROI区域采集过程。弹痕识别是一种精度很高的痕迹识别技术，对弹痕识别技术的深入研究是十分必要的，但弹痕识别、比对技术在我国还处于初级阶段，对其理论和应用的研究还需进一步深入和完善。因此，本文以高精度的弹痕图像为基础，针对弹痕识别中的关键技术问题进行了深入研究，在2D+3D图像融合、弹头/弹壳痕迹特征提取、基于多特征融合的弹痕识别及弹痕图像的检索等关键技术方面提出了有效的算法。本文首先分析了三维弹痕形貌数据的真实性，同时考虑到二维弹痕图像中也包含了大量重要的纹理信息，为提高识别率，并且保证客观性和可靠性，结合实际需要，提出了基于多尺度contourlet变换的2D弹痕图像和3D弹痕高斯曲率图像融合策略。在此基础上，针对弹头痕迹特点，提出了基于AFRAT弹头主痕迹特征提取算法、基于可控滤波器的弹头痕迹方向特征及基于二维log-Gabor小波变换的弹头纹理特征提取算法。同样，根据弹壳痕迹特征，将撞针痕迹与弹底窝痕迹分离，并讨论撞针痕迹形状特征提取方法及基于PCNN算法的弹底窝纹理特征提取方法。在分别对弹头痕迹和弹壳痕迹进行特征提取后，本文进一步研究了常见的特征匹配方法，较之传统的单一特征比对方法中识别率低的问题，本文提出了一种基于多特征融合的识别检索方法，通过多特征构成的相互独立的特征集，利用匹配级融合决策，构建了基于弹痕图像内容的多特征识别检索模型，通过实验证明，取得了理想的识别效果。在弹痕识别模型的基础上，本文还讨论了基于多级检索策略的弹痕图像检索方法，并将识别出的弹痕图像按照相似度降序排列，反馈给用户。最后本文设计可视化实验平台，并通过多个测试实验的分析、讨论，验证了算法和实验平台系统的可行性和有效性。
其他摘要	Ballistics comparison based on Firearms marks’is an important research area about the technology of the Criminal Investigation, making use of bullet trail inspection theory to judge and find out the emission guns by identifying the bullet trail, and provide important evidence of gun-related cases for the investigation process. At the current of our country, using microscope to observe and compare is the major method, but this inefficient and low accuracy rate approach can’t meet the bullet identification efficiency and accuracy needs. With the development of computer technology, the automatic bullet trail recognition technology is an effective way to identify the bullet trail. Bullet trail recognition technology is which make use of inherent trail features of bullet to identify the emission guns by digital image processing and pattern recognition theory. Since the importance and practicality of bullet trail recognition technology, domestic and foreign scholars have attached more importance to it. Bullet trail comparison technology has developed in recent years in our country, compared to the developed countries, our research is all based on the two dimensional gray-scale image, due to the two dimensional gray-scale images are influenced hardly by light intensity, angle and other environmental factors in the process of data acquisition, and the two dimensional image can’t reflect the topography information of bullet trail, so, it can’t guarantee the reliability of the identification result and the final discrimination of two bullets need to identify manually, the identification efficiency doesn’t improve essentially. And through the related fields study of other countries, the depth of the surface texture information of bullet is an important object of bullet trail identification research, so, according to the bullet trail comparison technical requirements, that ensure high accuracy while avoiding to damage the texture trail of bullets again. In this dissertation, a three-dimensional non-contact optical measurement system is adopted to acquire the information of bullets, and it can simultaneously obtains high precision three-dimensional topography data and two-dimensional gray-scale images, and introduces the principle, structure, and the calibration process of the device and bullet ROI regional acquisition process. Bullet trail recognition is a high precision trace recognition technology, and it is necessary to study on it deeply, but the bullet trail identification and matching technology in our country is still in its infancy, its theory and application should be research further and improvement. Therefore, the key technical issues of bullet identification are studied deeply in this dissertation based on high precision bullet trail images, and have proposed effective algorithms about 2D + 3D image fusion, feature extraction of bullet / cartridge, bullet trail identification and bullet trail images retrieval based on multi-feature fusion and other key technologies. In this paper，we analyze the authenticity of the three dimensional firearm marks topography data, and taking into account the two-dimensional image of the bullet also contains a mass of important texture information, combined with the actual needs, an image fusion strategy which fuses 2D ballistics image based on multi-scale Contourlet transform and 3D ballistics Gaussian curvature image is proposed to improve the recognition rate, and ensure the objectivity and reliability. On this basis, for the character of ballistics, we proposed main traces extraction algorithm of bus based on AFRAT and texture feature extraction algorithm based on two-dimensional Log-Gabor and Wavelet transform. Similarly, separate the trail of firing pin and breech face marks according to the characteristics of cartridge trail, and discuss the shape feature extraction of firing pin trail and texture feature extraction of breech face marks based on PCNN algorithm. After extract the features of ballistics independently, for issue of low recognition rate of traditional signal feature, propose a recognition retrieval method based on multi-features fusion through the study of traditional feature matching algorithms, In this paper ,we propose a retrieval method based on multi feature fusion, independent features formed by multiple sets of features, using the matching level fusion decision, based on multi feature recognition bullet image content retrieval model, proved by experiment, achieved satisfactory identification results. Based on the ballistics recognition model, we research the multistage retrieval method based on the strategy of the bullet image, and the bullet image identified according to the similarity in descending order, feedback to the user.Finally, we design visualization experiment platform, and through the analysis and discussions of multiple experiments, verified the feasibility and effectiveness of the algorithm and experimental platform systems.
语种	中文
文献类型	学位论文
条目标识符	http://ir.ciomp.ac.cn/handle/181722/41435
专题	中科院长春光机所知识产出
推荐引用方式 GB/T 7714	凌剑勇. 三维枪弹痕迹自动识别系统关键技术研究[D]. 中国科学院大学,2014.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
凌剑勇.pdf（9439KB）			开放获取	CC BY-NC-SA	请求全文