Study of binocular parallax estimation algorithms with different focal lengths

Existing binocular parallax estimation algorithms all use standard stereo cameras, which are unable to achieve high definition for monitoring both near and far objects due to the camera's own parameters. In order to solve this problem, a parallax estimation method is proposed for images taken by cameras with different focal lengths, which can directly calculate parallax maps for near and far views. A calibrated stereo correction method is proposed to parameterise the binocular image scaling for direct image correction at different focal lengths; the monocular portion of the results is calibrated by comparing the parallax results in the binocular range and introducing monocular parallax correction parameters. Experimental results on the Middlebury Stereo public dataset, which was used to simulate images of different focal lengths, show that the proposed parallax estimation algorithm was able to generate both far-field and near-field parallax maps with an avgerr of 8.33%, which is significantly better than the monocular parallax estimation method.