A universal vision-based roll angle estimation method for mobile robots

In this paper, a novel vision-based method for roll angle estimation of the robot is proposed. Inspired by biological visual cortex, scales and shapes of local patterns whose directions may contribute to the judgment of the scenes´ gravity and horizon are determined. On this basis, a local direction detector is designed and all the detected local directions in an image are summed up into the direction function. Then the precise directions of gravity and horizon may be easily obtained by using statistical techniques. Experiments illustrate the generalization and effectiveness of the proposed method.