基于车载机器视觉的安全带识别方法研究

发布时间：2019-06-17 14:56

【摘要】：为提高安全带佩戴率,论文从三点式安全带使用过程中存在的问题出发,分析三点式安全带的不同佩戴方式对乘员损伤的影响,利用车载机器视觉对其进行识别,提出安全带识别评价方法,构建满足实时性和高精度要求的识别模型,实现了嵌入式车载安全带监控系统设计。论文研究内容包括： 1)三点式安全带使用情况调查与分析。针对我国典型城市安全带的使用状况展开调查,结果显示,安全带使用过程中存在两类问题：不规范佩戴安全带和卷曲佩戴安全带。其中,导致安全带佩戴提示系统失效的不规范佩戴行为包括：单独使用安全带带扣、预佩戴和仅系肩带；卷曲佩戴包括：肩带卷曲、腰带卷曲和严重卷曲。 2)三点式安全带的不同佩戴方式对乘员损伤影响。采用MADYMO构建乘员约束系统模型,并验证模型的有效性。运用该模型进行三点式安全带在不同佩戴方式下的乘员损伤分析,结果表明：车辆碰撞时,预佩戴安全带的乘员将从座椅位置飞出；仅系肩带将产生明显的下潜、滑移运动。上述情况均导致乘员受到致命性损伤。相比正确使用三点式安全带,卷曲佩戴将导致乘员各项损伤指标明显上升。因此,对三点式安全带不同的佩戴方式进行识别,为充分发挥安全带应有的保护作用,具有积极意义。 3)车载视频监控平台搭建及安全带图像采集试验设计。搭建安全带图像采集试验平台,在CCD传感器性能参数、红外补光装置和特殊材料安全带等方面展开具体研究；设计并实施不同光线环境下的车辆行驶试验,采集获得安全带不同佩戴方式下的乘员图像信息；研究适用于安全带识别的图像预处理技术,为构建安全带识别的多特征参数模型奠定基础。 4)构建满足实时性要求的安全带识别模型。针对安全带在线检测的要求,采用主成分分析法降维后的安全带空间参数特征作为输入向量,选用BP神经网络作为分类器,构建基于BP神经网络的安全带识别模型,以满足实时性要求。引入遗传算法(GA)对其内部参数进行优化,构建基于GA-BP神经网络的安全带识别实时性模型,以满足准确性要求。通过硬件在环测试(HIL)和模型在环测试(MIL)验证了模型的实时性和准确性。 5)构建满足高精度要求的卷曲佩戴识别模型。提取安全带结构参数的统计特征值作为输入向量,选用支持向量机(SVM)作为模型核心分类器,以交叉验证方法对内部参数进行选择,采用粒子群算法(PSO)对其进行优化,构建基于PSO-SVM的高精度识别模型,并进行软件在环测试以验证代码的有效性,从而将其应用于高精度离线检测之中。 6)实现了嵌入式车载安全带监控系统的设计。从分析车载系统对软硬件的性能要求着手,探讨了DSP内核高速处理数据的特点以及ARM内核控制和管理的功能,选择ICETEK-DM642-B评估板作为硬件平台,实现了基于嵌入式技术的安全带识别系统的设计；完成系统功能的总体设计和功能模块的软件设计,并进行程序优化。论文研究的创新点如下： 1)提出基于车载机器视觉识别安全带的方法,搭建嵌入式车载安全带识别系统平台； 2)揭示了三点式安全带的不同佩戴方式(不规范佩戴和卷曲佩戴)与乘员损伤之间的关系； 3)提出安全带识别的实时性和准确性评价方法； 4)构建基于GA-BP神经网络的安全带识别实时性模型； 5)构建基于PSO-SVM的卷曲佩戴识别高精度模型。
[Abstract]:In order to improve the wearing rate of the safety belt, the paper analyzes the influence of the different wearing modes of the three-point safety belt on the injury of the passenger from the problems existing in the use of the three-point safety belt, The identification model satisfying the real-time and high-precision requirements is constructed, and the design of the embedded in-vehicle safety belt monitoring system is realized. The research contents include: 1) Investigation and distribution of the use of three-point safety belt Analysis of the use situation of the typical urban safety belt in China, the result shows that there are two kinds of problems in the use of the safety belt: the safety belt and the curling and wearing safety are not regulated Belt. In which, the non-standard wearing behavior that results in a failure of the safety belt wear warning system includes the individual use of the belt buckle, the pre-wear and only the shoulder strap; the crimp wear includes a shoulder strap curl, a belt curl, and a serious roll (2) Different wearing modes of the three-point safety belt to the passenger's loss Impact of injury. Build an occupant restraint system model using MADYMO and verify the model Effectiveness. The model is used to analyze the occupant injury of the three-point safety belt in different wearing modes. The results show that when the vehicle is in collision, the occupant of the pre-wear safety belt will fly out of the seat position; only the shoulder belt will produce a clear subpotential and slide. Moving. This results in a fatal occupant. Sexual damage. The use of a three-point seat belt in comparison to the correct use of a three-point seat belt will result in the occupant's various injury indicators In order to give full play to the protective effect of the safety belt, the wearing mode of the three-point safety belt shall be recognized and the product shall have the product. (3) Construction of on-board video monitoring platform and image collection of safety belt Set the test design. Set up the safety belt image acquisition test platform, and carry out specific research on the performance parameters of the CCD sensor, the infrared light-supplementing device and the special material safety belt, and design and implement the vehicle running test under different light environment, and collect the multiplication of the different wearing modes of the safety belt The paper studies the image pre-processing technology which is suitable for the identification of the safety belt, and provides a multi-feature parameter model for the safety belt recognition. form the foundation.4) Build the foundation to meet the real-time requirements According to the requirement of the on-line detection of the safety belt, the characteristic of the seat belt space parameter after the dimension reduction is taken as an input vector by adopting a main component analysis method, a BP neural network is used as a classifier, a safety belt identification model based on the BP neural network is constructed, The real-time requirement is met. Genetic algorithm (GA) is introduced to optimize the internal parameters, and the real-time model of safety belt recognition based on GA-BP neural network is constructed. The accuracy requirements are met. The model is validated by hardware in the loop test (HIL) and the model in the loop test (MIL) real-time and accuracy.5) The construction of high-precision requirements According to the method, a support vector machine (SVM) is used as the core classifier of the model, the internal parameters are selected by a cross-verification method, the particle swarm optimization (PSO) is adopted to optimize the internal parameters, and the PSO-SV is constructed. M's high-precision identification model, and the software is used in the loop test to verify the validity of the code, so as to apply it to high-precision off-line detection The design of the safety belt monitoring system is carried out. The characteristics of the high-speed processing data of the DSP core and the function of ARM core control and management are discussed from the analysis of the performance requirements of the on-board system. The ICETEK-DM642-B evaluation board is selected as the hardware platform, and the embedded technology is realized. The design of the safety belt recognition system; the overall design of the system function and the soft of the functional module The design of the part and the optimization of the program The innovation points of the thesis are as follows:1) Put forward the method to recognize the safety belt based on the on-board machine vision, and Built-in vehicle-mounted safety belt identification system platform;2) The different wearing modes of the three-point safety belt are revealed (not to be standardized the relationship between the wearing and curling of the wearer) and the occupant's injury;3) The real-time and accuracy evaluation method of safety belt recognition is put forward; and 4) the basis of construction Real-time model of safety belt recognition based on GA-BP neural network
【学位授予单位】：江苏大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：U463.6;U491.61

【参考文献】