智能交通中行人检测算法的研究与实现

发布时间：2018-01-13 22:28

本文关键词：智能交通中行人检测算法的研究与实现　出处：《大连海事大学》2014年硕士论文　论文类型：学位论文

【摘要】：在智能交通领域中,行人的检测是一个重要且基本的任务,由于检测场景的复杂性,以及行人本身形态、姿态的复杂多变性,使得行人检测,特别是基于嵌入式设备的行人检测一直是计算机视觉研究中颇具难度的问题。随着智能交通系统的发展,行人检测系统有着更广阔的应用前景,但同时也对系统的体积、功耗、成本及稳定性提出更高的要求,研究基于嵌入式的行人检测系统具有重大的实用价值。本文设计实现一个基于Adaboost学习算法并且在TMS320DM642平台上实现的行人检测系统。论文的主要内容和研究成果包括： 1.根据对行人检测算法的相关理论的学习,选择了可移植到TSM320DM642平台上的基于Adaboost的行人检测算法,本文设计的系统使用基本的Haar-like矩形特征以及专门应用于行人检测三角特征,应用积分图像的方法计算特征值。利用OpenCV实现了基于Adaboost的行人检测算法的仿真,训练得到级联分类器,在PC机上进行了多组实验,取得良好的检测效果。 2.根据设计需要,对硬件系统进行选择规划,选择TI公司的TMS320DM642芯片作为主控芯片,同时对TMS320DM642视频口的配置进行细致的研究学习,选择视频采集输出芯片,并根据PCB设计原则设计视频版的PCB板。 3.利用集成实时操作系统DSP/BIOS进行软件开发,在RF5框架下,将系统分为图像采集、图像处理和图像显示三个部分。并将基于Adaboost的行人检测算法移植到DM642平台上,利用CCS提供的各类调试工具对代码进行分析优化,通过实验测试表明基于DM642的行人检测系统每秒可检测10帧图像,基本满足实时性要求。
[Abstract]:Pedestrian detection is an important and basic task in the field of intelligent transportation. Due to the complexity of detection scene and the complex variability of pedestrian morphology and posture pedestrian detection makes pedestrian detection. In particular, pedestrian detection based on embedded devices has been a difficult problem in computer vision research. With the development of intelligent transportation system, pedestrian detection system has a broader application prospects. But at the same time, the volume, power consumption, cost and stability of the system are also required. It is of great practical value to study the pedestrian detection system based on embedded system. This paper designs and implements a learning algorithm based on Adaboost and realizes pedestrian detection on TMS320DM642 platform. The main contents and research results include:. 1. According to the theory of pedestrian detection algorithm, we choose the pedestrian detection algorithm based on Adaboost which can be transplanted to TSM320DM642 platform. The system designed in this paper uses the basic Haar-like rectangular feature and is specially applied to the pedestrian detection triangle feature. The method of integral image is used to calculate the eigenvalue. The simulation of pedestrian detection algorithm based on Adaboost is realized by using OpenCV. The cascade classifier is trained and many experiments are carried out on PC. Good results were obtained. 2. According to the design needs, the hardware system is selected and the TMS320DM642 chip of TI company is chosen as the main control chip. At the same time, the configuration of TMS320DM642 video port is studied carefully, the video capture and output chip is selected, and the PCB board of video version is designed according to the design principle of PCB. 3. The software is developed by using DSP/BIOS, which is an integrated real-time operating system. The system is divided into image collection under the framework of RF5. Image processing and image display are three parts. The pedestrian detection algorithm based on Adaboost is transplanted to DM642 platform, and the code is analyzed and optimized by various debugging tools provided by CCS. The experimental results show that the pedestrian detection system based on DM642 can detect 10 frames of images per second, which basically meets the real-time requirements.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;U491.226;TP391.41

【参考文献】