Binghua Guo, Hongyue Dai, and Zhonghua Li
Visual saliency, motion detection, dynamics, mobile robots,navigation
Saliency is an important part of the human vision mechanism.
A visual saliency computing model that imitates this mechanism
makes mobile robots more eﬃcient and robust in motion detection.
A method of motion detection based on visual saliency for mobile
robots is presented in this paper. First, the matching SURF features
in two consecutive frames are extracted, and 3D projection geometry
theory and the multi-RANSAC algorithm are utilized to estimate
the velocity of the dynamic objects. Second, a potential-gradient
function is deﬁned as motion contrast to determine the saliency
of dynamic objects. The saliency of objects makes mobile robots
suitable for the detection of high-dynamic or surrounding moving
objects. Lastly, integration is demonstrated with visual saliency
modelling in both indoor and outdoor environments. Results of the
experiments and evaluation revealed the capabilities of the proposed
approach for motion detection. As a result, the proposed scheme
can provide an eﬃcient processing method to meet the requirements
of moving object detection for robots.