基于计算机视觉的复杂背景人行桥振动识别

Computer vision-based vibration recognition of pedestrian bridges with complex backgrounds

基于振动的桥梁健康监测是以定量方式评估桥梁安全性能的有效工具,针对在使用传统接触式传感器时,实现大型结构的高空间分辨率测量难以达成,并且固定接触式传感器在超过主体结构的寿命时还可能缺乏可靠性。文章建立一套非接触桥梁振动识别系统,系统对基于相位的运动估计(PME)进行改进,结合2DGabor小波并抽出相位中与积分无关的变量,提出新的表示方法用于提取相位信息。系统利用改进的PME对结构表面的自然目标进行追踪,进而识别全场结构的振动信息;再通过模态分解离散出结构有效振动信号,将高频微小信号通过基于相位的运动放大(PMM)技术进行放大,最终识别出完整的结构振动信息。为应对城市复杂环境下桥梁振动识别的困难,消除环境对相机振动的影响,文中提出利用模态分解剔除环境对相机振动的影响和其他噪声,进而提取出有效的桥梁振动信息。通过在实验室桥梁模型上进行三种不同工况的试验和户外人行桥斜拉杆测试,验证系统对桥梁振动识别的可行性。在模型试验中,系统识别得到的时域和频域信息与激光位移传感器对比的误差都小于0.80%;户外人行桥斜拉杆测试中,系统识别的结果与接触式高频加速度传感器对比的误差小于2.0%,比桥梁挠度仪的可靠性更高,且利用模态分解离散出拉杆的各阶瞬时频率,由此表明系统鲁棒性强且经济性好,具备广泛的应用前景。

Vibration-based health monitoring of bridge is an effective tool for quantitatively assessing bridge safety performance,against the facts that high spatial resolution measurements of large structures is generally difficult to achieve using conventional contact sensors and that fixed contact sensors may also lack reliability when the lifetime of the main structure is exceeded.In this work,a non-contact vibration identification system of bridge is developed.In the system,the phase-based motion estimation(PME)is improved by combining 2D Gabor wavelets and extracting the integralindependent variables in the phase,and a new representation is proposed for extracting the phase information.The system uses the improved PME to track the natural targets on the surface of the structure,and then identifies the vibration information of the full-field structure;then the effective vibration signal of the structure is discrete by mode decomposition,and the high-frequency tiny signal is magnified by the phase-based motion magnification(PMM)technique to finally identify the complete structural vibration information.To cope with the difficulties of bridge vibration identification in an urban complex environment and eliminate the influence of environment on camera vibration,it is proposed to use mode decomposition to eliminate the influence of environment on camera vibration and other noises,and then then extract the effective bridge vibration information.The feasibility of the system for bridge vibration identification is verified by conducting tests on a laboratory bridge model with three different working conditions and by outdoor pedestrian bridge tilt-tie tests.In the model tests,the errors of the time and frequency domain information obtained by the system are less than 0.80%compared with those of the laser displacement sensor;in the outdoor pedestrian bridge tilt-tie tests,the errors of the system are less than 2.0%compared with those of the contact high-frequency acceleration sensor,which is more reliable than the bridge deflection meter.Hence,it is shown that the system is robust and economical,and has a wide range of application prospects.