Research On 3D Thermography Reconstruction Based On Depth Sensor

Infrared thermography provides a passive and noncontact tool to reflects the surface temperature distribution of target objects by means of 2D images.It has been applied in a broad range of applications such as medical diagnosis,fault detection,energy assessment and intelligent driving.However,using 2D infrared image to express the surface temperature distribution of an object has the following severe limitations:a.it can only obtain temperature information of an object at a single view,b.lacks 3D information of the temperature anomaly area,and can't accurately locate the hot spot position,c.the dimensionality reduction mapping in the image acquisition process leads to information loss and low temperature measurement accuracy.Therefore,this paper proposes a 3D thermography reconstruction technique based on depth sensor.Firstly,by analyzing the working principle of the depth camera and the imaging principle of the long-wave thermal infrared camera,the calibration checkerboard which can be simultaneously observed by the depth camera and the thermal camera is designed to complete the internal and external calibration of the depth camera and thermal camera.Next,according to the low texture of thermal images and thermal information stable to the observed angle and environmental factors,based on ICP algorithm,we proposed the Thermal-guided ICP algorithm to match the current data with the model.Then,we observed the matching effect between the depth information and thermal information by using the calibrated external parameters in different scanning situations,and proposed the external parameters compensation strategy based on Thermal-guided ICP to solve the depth map and thermal map deviation of external parameters caused by different moments of capturing.Finally,based on the TSDF model,the extended TSDF model is constructed,and a reasonable and effective data fusion strategy is proposed based on the depth map characteristics and thermal information influencing factors.The main contents include:Chapter 1 analyzes the status of the 3D thermography reconstruction technology.The main difficulties faced by 3D thermography reconstruction are summarized.The research content of this paper is given.Chapter 2 describes the hardware components of the 3D thermography reconstruction system and the calibration of the depth and thermal cameras.This chapter first introduces the depth and thermal cameras type and parameters used in the 3D thermography reconstruction system.Then,internal parameters of the depth and thermal cameras are separately calibrated by a special calibration checkerboard.Finally,the external parameters between depth and thermal cameras is initialized by plane matching and then refined by point matching.Chapter 3 studies The multi-view matching algorithm of 3D thermography reconstruction system to estimate the precise pose of the camera.Firstly,the overall framework of Thermal-guided ICP is expounded.Then,the image matching of the front and rear thermal maps is used to obtain the reasonable rough matching point pairs.And SVD is used to estimate the initial relative pose between the two frames.Finally,the initial pose is refined by maximizing geometric and thermal information consistency between the frame and the model.Chapter 4 studies the simultaneous optimization of camera pose and external parameter deviation between cameras,based on Thermal-guided ICP algorithm.Based on the clock synchronization and timestamp nearest neighbor strategy,combined with the Thermal-guided ICP algorithm to maximize the thermal information consistency between the thermal map and the model,solves the external parameter deviation between the cameras.Chapter 5 studies multi-view data fusion strategies.Firstly,the TSDF model is extended to a capable of storing geometry and thermal information TSDF model.Then,based on the depth map characteristics,the distance of capturing and the relationship between the thermal information and the emission angle,an effective data fusion strategy is proposed.Finally,the mesh model is extracted using Marching Cubes.Chapter 6 summarizes the work of this paper and looks forward to the future work.