Research On Key Technologies Of 3D Reconstruction On Mobile Platforms

3D reconstruction is a hot research area in computer vision field.The structured light based 3D reconstruction method can achieve a high resolution reconstruction result,but the equipment are complex and the costs are high.The stereo image based technique has solved the problem of high cost,however,the poor portability and the complex calibration also limit its usable range.The mobile platform based equipment are integrated with multiple sensors,and the power of computation has increased steadily,how to realize 3D reconstruction based on mobile platform has become a hot research field.The key technologies in 3D reconstruction using single camera are how to steadily obtain the camera's extrinsic parameter.This paper's work focus on how to obtain camera's extrinsic parameter using the IMU(inertial measurements units)combined with the captured images.The main works and achievements are as follows:1.This paper proposes an Extended Kalman Filter to integrate camera positions calculated from IMU and from images.To solve the problem of scale drift in loop closure situations,the Extended Kalman Filter has taken into account the similarity transforms to achieve a better performance.Experimental results show that by using this strategy,we can achieve a more accurate camera position than using IMU only or image data only.Since IMU can measure camera's movements in physical scale,we can obtain the 3D object's physical size,it is an advantage over image based 3D reconstruction.2.To achieve a more accurate initialization results,this paper propose an initialization algorithm using image data and IMU data.The algorithm first computes camera positions from IMU data and from image data,the most accurate one was then selected to perform the initialization step.Experiments show using this algorithm the camera position is more accurate than algorithms using image data only.3.Under some circumstances such as a quick movement of camera,the vision tracking might get lost and the system is unable to achieve the extrinsic parameter.We propose a quick relocalization strategy using the IMU sensor.The system computes the position similarity of the current image frame and the successfully tracked frames,and chose the most similar one to solve the PnP problem.4.Finally we built a 3D reconstruction system based on the proposed methods.We use the indoor scenes to test the system and achieve good results.