The3D Reconstruction Technology Study Of Moving Object Based On Multi-sensor Information Fusion

With the continuous development of sensor,the information of different sensor inthe same space is different. Through the combination of different sensors,complementarily between each sensor has also received widespread attention.This paper studied dater fusion of laser sensor, visible light image sensor andinfrared image sensor, also studied the detection and3D reconstruction of the targetobject. This paper mainly includes data pretreatment of built real time laser acquisitionplatform, external setting of laser and camera, the fusion of visible image and infraredimagery, detection of moving object and extraction of its point cloud, rebuilding andcolor mapping of three dimensional gridding model.Write the real time acquisition platform of the laser data. Do pretreatment of laserdata using filtering method, filtering bulky noise for reduce error. Using calibrationcube for calibration mediation, take the relationship of linear equations of three edgesof the cube in the image coordinate and the laser points on the three edges as constraintcondition, for realize external calibration of laser and camera.Implement optical and infrared image transforming fusion based on NSCT.Decompositing image with non-sampling Countourlet conversion, get high frequencysubband images and low frequency subband, add high frequency subband with borderdetection, get new high frequency subband, fusion high frequency and low frequencysubband of the source image with respective fuse rule,rebuilding inversetransformation using non-sampling Contourlet conversion, getting the final fusionimage.Reconstruct the3D object by processing point cloud data of the region of interestsegmented from infrared image. Firstly, segment the moving object from the infraredimage using optical flow method; then extract the objective laser point cloud data withthe calibration result of laser and camera; next build the3D gridding model of theobject by Delaunay triangularization method. Finally, map the objective color value infusion image onto the extracted point cloud data by the calibration, making the3Dreconstruction of the object more visual.