基于联合自适应卡尔曼滤波的缩微车目标检测系统设计与实现

发布时间：2018-01-31 09:48

本文关键词： 目标检测运动状态估计多传感器缩微车行人识别联合自适应卡尔曼滤波　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：车辆的目标检测主要是通过车载传感器对局部环境感知来实现,其中最重要的部分是估计车辆前方目标的运动状况。车辆行驶过程中对前方目标运动状态的估计是智能车辅助驾驶系统中的重要研究课题,其主要包括实时测量搭载传感器的实验车与前方目标的相对横纵向速度、相对横纵向加速度和相对位置等状态变量。传统的单传感器目标检测只能根据传感器的特性获取某一方面的数据,而多传感器可以为缩微车提供更加完整的数据信息。因此可以将自适应卡尔曼滤波模型和联合卡尔曼滤波模型做出融合,达到了避免单传感器信息采集盲区的问题并且提高了滤波器系统稳定性的目的,最终达到提高多数据源信息融合的质量目的,为进一步的判断和决策提供了良好的基础。另一方面采用缩微车技术可以降低实验成本、降低实验风险、提高系统维护性并且提供了模拟大规模车辆的功能,这些特点使得缩微车成为了智能交通实验的良好载体。本文以场景中的缩微车为基础,即每个车辆都可以获得一定范围内的其他车辆和障碍物的状态信息,通过以下两个方面对该系统进行设计与实现:(1)实现对缩微车前方目标检测:缩微车在行驶过程中通过摄像头、雷达和车载光电编码器不断采集车辆前方的局部交通情况,然后通过mini2440处理器对来自不同传感器的信息进行集中处理。在mini2440中将进行如下算法:采用基于HOG+SVM行人识别提取图像数据的行人目标、采用区域分割法提取雷达数据的目标运动信息以及通过光电编码器获取车辆自身车速。最后经过处理的信息将发送到服务器端并进行数据融合完成局部交通环境的显示。(2)实现多传感器数据融合:基于新息自适应卡尔曼滤波,提出了一种多传感器联合滤波的方法来提高估计前方行驶车辆的运动状态的精度。该滤波方式以激光雷达和毫米波雷达作为局部滤波器,通过自适应过程实时调整局部滤波器的测量噪声协方差和系统噪声协方差,并利用测量噪声协方差计算出联合滤波中各局部滤波器的信息分配因子,最后根据信息分配因子来完成最优信息融合并将全局滤波结果反馈到各局部滤波器。联合卡尔曼滤波可以结合多种传感器的测量优势,使每次得到的滤波结果是各个局部滤波器中最优的,并且利用反馈机制提高了测量精度。根据实验结果可知,本文提出的缩微车目标检测系统的功能正确完善,可以达到实时监控局部交通环境的目的。同时联合自适应卡尔曼滤波方式有良好的环境适应能力和系统容错率,当一个传感器的滤波性能出现异常时,该系统的滤波结果并未产生异常。在整个实验过程中,该方法相较于标准卡尔曼滤波和单传感器的新息自适应滤波具有最高的测量精度。
[Abstract]:The target detection of vehicle is mainly realized by the local environment perception of the vehicle sensor. The most important part is to estimate the moving state of the vehicle in front of the vehicle. The estimation of the moving state of the forward target is an important research topic in the intelligent vehicle assisted driving system. It mainly includes the real-time measurement of the relative transverse and longitudinal velocities between the experimental vehicle and the target in front of the sensor. The traditional single sensor target detection can only obtain some data according to the characteristics of the sensor. The multi-sensor can provide more complete data information for the minibus, so the adaptive Kalman filter model and the United Kalman filter model can be fused. It can avoid the problem of single sensor information acquisition blind area and improve the stability of filter system. Finally, it can improve the quality of multi-data source information fusion. On the other hand, it can reduce the cost of experiment, reduce the risk of experiment, improve system maintainability and provide the function of simulating large-scale vehicle. These characteristics make the minibus a good carrier of intelligent traffic experiment. This paper is based on the scene of the minivan, that is, each vehicle can obtain a certain range of other vehicles and obstacles of the state information. Through the following two aspects of the design and implementation of the system: 1) to achieve the microcar forward target detection: the minibus in the process of driving through the camera. Radar and on-board photoelectric encoder continuously collect the local traffic information in front of the vehicle. Then the information from different sensors is processed centrally through the mini2440 processor. In mini2440, the following algorithm will be used: adopt HOG based on. The pedestrian target extracted from the image data is identified by SVM pedestrian recognition. The region segmentation method is used to extract the target motion information from radar data and the vehicle speed is obtained by photoelectric encoder. Finally, the processed information will be sent to the server and fused to complete the local traffic environment. Display of. 2) realize multi-sensor data fusion: adaptive Kalman filter based on innovation. A multi-sensor combined filtering method is proposed to improve the accuracy of estimating the moving state of the moving vehicle in front of the vehicle. The filtering method uses laser radar and millimeter wave radar as local filters. The measurement noise covariance and the system noise covariance of the local filter are adjusted in real time by the adaptive process, and the information allocation factor of each local filter in the joint filter is calculated by using the measurement noise covariance. Finally, according to the information allocation factor, the optimal information fusion is completed and the global filtering results are fed back to the local filters. The United Kalman filter can combine the measurement advantages of various sensors. The filter results obtained each time are the best in each local filter, and the measurement accuracy is improved by the feedback mechanism. According to the experimental results, the function of the microbus target detection system proposed in this paper is correct and perfect. At the same time, the combined adaptive Kalman filter has good environmental adaptability and system fault tolerance, when the filtering performance of a sensor is abnormal. In the whole experiment, this method has the highest measurement accuracy compared with the standard Kalman filter and the single sensor innovation adaptive filter.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TN713

【相似文献】