Supporting drivable region detection by minimising salient pixels generated through robot sensors

Abstract

The role of robots, automatically guided machines able to perform tasks on their own cannot be over emphasized. In particular, if robotic vehicles are to work effectively, the way they are required to perform their jobs and their ability to reach the desired destination where the job is to be performed are of utmost importance. This necessitates the need to facilitate proper navigational aid for robotic vehicles. Various navigational approaches have been proposed in robotics literature, but this work serves to provide an assistive pre-processing strategy for the detection of drivable region through minimisation of salient pixels in a colour feature extraction. Salient pixels are pixels occupying the non-drivable region particularly those having same grayscale value as road images. Salient pixels provide difficulties during colour feature extraction on road images captured by a robot’s camera (sensor). In our method, a stream of road images is captured, pixels are extracted based on a RGB (red, green, blue) colour space, edges of objects are detected using Sobel operator. Salient pixels are minimised using some heuristic which is based on a threshold parameter. In a series of experiments using our method, a stream of real life road images is obtained and results show that good drivable regions, which facilitate proper robotic navigation, can be detected