Fast stereo matching using adaptive stripe-based optimization

Stereo-based advanced driver assistance systems have become more popular because stereovision enables to obtain a three-dimensional (3-D) representation of the environment around a vehicle. However, such information may not always be accurate because the reference images differ in exposure level and illumination. In addition to this, most current implementation of ADAS systems benefits from hardware devices and graphics processing units in order to achieve high frame rates. To overcome these issues, we propose a Census-based stereo matching technique which uses adaptive length optimization paths via multiple stripes for finding the most accurate depth value. Experimental results show that the proposed scheme outperforms the reference methods by reducing the overall time complexity and improving the quality of the estimated depth maps.