A Multi-Sensor Information Fusion Approach for Efficient 3D Reconstruction in Smart Phone

Despite the recent advancements, camera pose estimation using smart phone’s sensor data is still errorprone due to various environmental noise and variability of the force applied on the phone during data acquisition. Here, we propose a novel framework to mitigate these drawbacks of camera pose estimation using various sensor’s data. The directional information of accelerometer is utilized to obtain reliable features for position estimation, rather than using its magnitude. Moreover, a multi-sensor data fusion approach is followed for robust rotation estimation. The proposed framework improves the epipolar geometry estimation and produces accurate stereo-correspondence points. This in turn reduces the reprojection error and enhances the quality of 3D reconstruction. Furthermore, the proposed framework utilizes only the CPU of the phone, instead of GPU and takes around 2-3 secs for processing an image. Extensive experimental results show supportive evidence of the effectiveness of the proposed framework in pose estimation problem.