Fig. 1. Schematic diagram for camera pose estimation with unknown focal length

Fig. 2. Comparison of camera pose and focal length errors in synthetic data experiments at noise intensity σ=2, 5, 10, 15 pixel. (a)‒(d) Rotation matrix errors; (e)‒(h) translation vector errors; (i)‒(l) focal length fu error of PnPf algorithm; (m)‒(p) focal length fv error of PnPf algorithm

Fig. 3. Four images from ETH3D Benchmark dataset, with the first two being outdoor scenes and the latter two being indoor scenes. (a) Terrace; (b) courtyard; (c) terrains; (d) relief

Fig. 4. Comparison of camera pose and focal length errors, as well as CPU running time in real data experiments. (a)‒(d) Rotation matrix errors in terrace, courtyard, terrain, and relief; (e)‒(h) translation vector errors in terrace, courtyard, terrain, and relief; (i)‒(l) CPU running time in terrace, courtyard, terrain, and relief; (m)‒(p) focal length fu error of PnPf algorithm in terrace, courtyard, terrain, and relief; (q)‒(t) focal length fv error of PnPf algorithm in terrace, courtyard, terrain, and relief