Image Based Three Dimensional Measurement

When traffic police deal with accident, they must investigate and take photo of the accident scene, and then make traffic accident scene graph according to relative laws and regulations. This paper provides a convenient tool to measure scene within certain satisfactory degree of accuracy, and to measure the missed proof from photos if the scene no longer exists. It takes comparatively less time to take photo, which reduces the period of jamming due to accident, hence improving the efficiency of traffic accident handling.This thesis reconstructs point in Euclidean space based on the theory of projective geometry, epipolar geometry, etc. First we calibrate camera and get a pair of projective camera matrix from fundamental matrix computed with assigned corresponding image points. Based on the hypothesis that some property of calibrated matrix is known and some is constant, we derive some constrains on absolute dual conic in the projective space, and then compute the transformation H, updating this projective space into a similar one. On the other hand, we optimize the image point pairs to make them fit for epipolar constrains, then with the pair of projective camera matrix above, we get their projective position and finally update these projective ones into similarity ones with H. Because our experiments use two photos, which does not provide enough constrains to determine a unique absolute conic, Cheirality constrains thus are used to filter those unreasonable. Then if the length of one line segment is known, we can get the real length of any other line segment according to their ratio of length in similar space.The main contribution of this paper is to use Cheirality conditions instead of special scene content to confine the Absolute Dual Conic, thus spreading the sphere of application. Nowadays, some digital cameras are capable of changing focus length automatically, and our method guarantees that this has no influence on measurement by not using focus length as its parameter. Experiment results show most of errors within 7%, thus satisfying certain demand. Besides, the camera used is daily digital camera, nice price and this reduces cost of measurement.