Efficient relative camera orientation detection for mobile applications

Nowadays, an intelligent mobile phone is required to have at least a camera and be able to process images with multiple sensors. However, when people take a photo at a random place, it is hard to build a whole pipeline to calculate everything accurately and efficiently. Therefore, basic and efficient algorithms are needed for both program developers and terminal users. In this paper, we will focus on a basic but important problem - relative camera orientation calculation, which is part of camera calibration. We introduce an efficient algorithm to tell the relative angle between the main object in the image and the camera. While no training data is used other than the test image itself, line segments are extracted by a Hough Transformation, based on Canny's edge detection algorithm. By extracting the focal length from the EXIF information and calculating the pixel lengths of some critical segment lines, the algorithm will give an accurate orientation for the main object (building) in the image. With the advantage of real time processing, our work can be further used for image location detection and other cute applications on intelligent mobile phones. Experiments show its effectiveness and efficiency on both experimental data of the campus and web data of some famous buildings.