基于DAVINIC技术的车载红外行人检测系统的研究和实现

发布时间：2018-06-24 14:19

本文选题：红外行人检测 + DM6437　；参考：《中国科学院大学（工程管理与信息技术学院）》2015年硕士论文

【摘要】：因为路灯和车灯光照距离有限,再加上驾驶员夜间开车容易疲劳,夜间行驶极易发生行人被撞的交通事故。车载红外行人检测作为车辆辅助驾驶的一部分,可以在较远的地方就提醒驾驶员行人的位置,从而提前做出判断,一定程度上能大大降低交通意外发生的概率。车载红外行人检测系统采用被动式红外相机获取外部图像,硬件平台基于TI公司的DAVINIC系列的芯片DM6437,软件使用BIOS操作系统,并实现红外行人检测算法在DM6437上的移植,基于DSP的内部系统流程,对算法进行最大化的优化,保证系统实时运行,基本满足实际使用需求。论文主要工作包括： (1)需求分析和硬件结构,以需求为基础分析红外车载行人检测系统的几种方案,比较各自优缺点,本系统的一个创新点是选用硬件DSP加软件算法的方案,然后分析主流DSP的优缺点,针对本系统的需要,选择DM6437做主芯片,硬件系统接口分析,分析了DM6437视频输入输出系统,按系统需求设计了视频编码模块接口,包括TVP5146的介绍和选择它的原因和其功能；外部存储模块DDR2接口分析,分析采用DDR2芯片的必要性和可行性,根据传输速度,大小等选择合适的DDR2接口的存储器,并实现该芯片和DM6437的接口；外扩FLASH存储的的必要性分析,DM6437的EMIF控制器以及其和具体FLASH芯片的接口,I2C/CAN接口分析。 (2)软件系统分析和设计,主要是实现整个系统软件开发实现：包括CCS3.3开发环境搭建；为CCS33添加VLIB库；软件实现的基本平台BIOS操作系统介绍并为本系统配置BIOS系统；通过BIOS提供的API函数实现软件架构搭建。介绍了红外行人检测算法的实现,包括红外行人分割、识别和行人跟踪算法。红外行人检测算法代码实现主要的流程包括读取一帧视频图像、阈值分割、区域处理、模板匹配等四个过程。在此软件架构的基础上移植红外行人检测算法,实现整个软件算法的流程,最后实现软件代码调试. (3)实现了红外行人检测算法在DM6437的优化,包括软件流水的原理；编译器的优化：通过提升代码局部性的优化；基于改变数据类型的优化；基于数值操作的优化；减少函数调用以及跳转指令的优化；采用内联函数的优化等等,每个优化方式的后面给出该优化方式对本论文算法的优化的效果,比较其差异性,最后列出了各种优化后的效果的表格,并展示了系统效果图。本系统的一个创新点是在分析了DSP运行机制的基础上在本系统上使用了所有的优化方式,对优化实现最大化。本论文讨论的红外视频检测系统,在整体识别过程中,使行人识别检测率达到90%左右,优化后在DM6437上的处理速度达到了20帧每秒,达到视频实时显示,故该系统具有较强的工程应用价值。
[Abstract]:Because the distance between street lights and vehicle lights is limited, and drivers are easily tired in driving at night, pedestrian accidents occur easily at night. As a part of vehicle auxiliary driving, infrared pedestrian detection can remind the driver of the pedestrian position in the distance, and make the judgment in advance, which can greatly reduce the probability of traffic accident to a certain extent. Vehicle infrared pedestrian detection system uses passive infrared camera to obtain external images. The hardware platform is based on TI's DAVINIC series chip DM6437.The software uses BIOS operating system, and realizes the transplantation of infrared pedestrian detection algorithm on DM6437. Based on the internal system flow of DSP, the algorithm is optimized to ensure the real-time operation of the system and meet the practical needs. The main work of this paper includes: (1) requirement analysis and hardware structure. Several schemes of infrared vehicle pedestrian detection system are analyzed based on requirements, and their advantages and disadvantages are compared. One innovation of this system is to choose the scheme of hardware DSP plus software algorithm, then analyze the advantages and disadvantages of mainstream DSP, select DM6437 as the main chip to meet the needs of the system, analyze the hardware system interface, and analyze the DM6437 video input and output system. The video coding module interface is designed according to the system requirements, including the introduction and selection of TVP5146 and its function, the external storage module DDR2 interface analysis, the necessity and feasibility of using DDR2 chip, according to the transmission speed, The memory of the appropriate DDR2 interface is selected, and the interface between the chip and DM6437 is realized. Analysis on the necessity of Expanding Flash Storage EMIF controller of DM6437 and its interface with flash chip I 2C / can interface analysis. (2) Software system analysis and design, mainly to realize the whole system software development and implementation, including CCS3.3 development environment; We add VLIB library for CCS33, introduce the basic platform BIOS operating system of software implementation and configure BIOS system for this system, and build the software architecture through API function provided by BIOS. This paper introduces the realization of red pedestrian detection algorithm, including infrared pedestrian segmentation, recognition and pedestrian tracking algorithm. The main process of code implementation includes reading a frame of video image, threshold segmentation, region processing and template matching. On the basis of this software architecture, the infrared pedestrian detection algorithm is transplanted to realize the flow of the whole software algorithm, and finally the software code debugging is realized. (3) the optimization of the red pedestrian detection algorithm in DM6437 is realized, including the principle of software pipeline; Compiler optimization: optimization by improving code locality; optimization based on changing data types; optimization based on numerical operations; optimization of reduced function calls and jump instructions; optimization of inline functions, etc. At the end of each optimization method, the effect of the optimization method on the algorithm of this paper is given, and its difference is compared. Finally, the tables of various optimized results are listed, and the system effect diagram is shown. One of the innovations of this system is that we use all the optimization methods on the basis of analyzing the DSP operation mechanism to maximize the optimization. The infrared video detection system discussed in this paper makes the detection rate of pedestrian recognition reach about 90% in the whole recognition process, and the processing speed on DM6437 reaches 20 frames per second, and the real-time video display is achieved. Therefore, the system has strong engineering application value.
【学位授予单位】：中国科学院大学（工程管理与信息技术学院）
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U463.6;TN219

【参考文献】