Abstract:

In this paper, we present a new approach for 3D reconstruction of agricultural field scene using a hand-hold digital camera. Firstly, EXIF information such as focal length was abstracted from image in JEPG format, camera intrinsic matrix could be solved based on these parameters. Camera extrinsic matrix was acquired from SVD decomposition of essential matrix which stand for matching relationship between two images, with intrinsic and extrinsic matrix camera projection matrix was available. Secondly, in order to generate dense 3D points cloud of field scene, each pixel was matched up according to SSD value, under the searching threshold from the differences of SIFT matching points in two images. Finally, the initial 3D model was refined with bundle adjustment, map the RGB value to correspondence 3D point, we got the final agricultural field scene. From the 3D reconstruction result, we can draw the conclusion that, our method has a good reflection of real world, and has the advantage of concision, and highly automated.