基于红外图像形心优化提取的汽车夜视安全系统  

Night Safety System on Vehicle Based on Optimization and Extraction of Infrared Image's Centroid

作　　者：吴琼飞[1] 朱勇[2] 张志强[1] 

机构地区：[1]华中农业大学楚天学院信息工程学院,武汉430205 [2]武汉纺织大学数学与计算机学院,武汉430200

出　　处：《计算机测量与控制》2015年第6期2132-2134,2138,共4页Computer Measurement &Control

基　　金：国家自然科学基金项目(31171694)

摘　　要：为了提高汽车夜间行驶安全性能,设计并实现了一套基于红外图像处理的夜视安全系统;使用被动式红外相机对车辆和行人进行红外成像以得到车辆前方行人和车辆的准确位置;测距必须确定对象的形心,对红外图像的灰度值进行判定区分行人和车辆,对行人使用中值空域滤波方法进行降噪,对车辆使用db4小波进行预处理,然后使用基于图像边缘的图像分割方法进行图像分割,使用最大类间方差评估确定分割结果,再进行区域提取得到形心坐标;使用小孔成像模型计算形心与本车的距离,提取本车速度、加速度参数计算理论碰撞时间,当理论碰撞时间小于时间阈值时进行声光告警,从而实现实时预警;系统使用DIR-384-P红外相机,系统算法使用PFGA实现,测试结果表明,经过标校和加固后,系统与目标存在5 m距离时,车辆测距的误差率达到2.74%,行人测距的误差率低于3.9%,具有一定的实际应用价值。In order to improve the safety performance of automobile driving at night, a night safety system based on infrared image pro cessing is designed and implemented. In the system the passive infrared camera is used to get infrared image so as to gain the accurate position of the front car or pedestrian. The location calculation need to determine the object centroid by the gray value to distinguish pedestrians or vehicles, after which the median spatial filtering method is used for denoising the infrared images of pedestrian and the db4 wavelet is used to preprocessing for vehicles, and the image segmentation method based on image edge is used and the final value is generated by the maximum between class variance evaluation. At last the centroid coordinate is obtained with the region extracting. The distance between the vehicle and the eentroid of the front one by the pinhole imaging model, at the same time the speed and acceleration parameters of the own car are extrac ted for computing the ideal collision time. When the ideal time is less than the time threshold the system will do acoustic--optical alarm so as to output the real--time warning. In the system the D1R--384--P infrared camera is included and the algorithm is realized by FPGA. The test results show that the error rate of vehicle distance measurement reached 2.74% and that of pedestrian distance measurement reached 3.9M in the distance range less than 5 meters after calibration and reinforcement, and there is a certain practical application value.

关 键 词：红外图像处理 汽车夜视安全系统 被动式红外相机 图像分割 碰撞时间 

分 类 号：U463.6[机械工程—车辆工程]