Visual Saliency under Uncertain Workspace of Assistive Robots

Robot navigation in uncertain environments has received relatively less attention due to the challenges posed by such environments. Examples of uncertain environments include navigating the robot around nuclear reactors, thermal captive power plants, aluminum smelters, steel plants, and chemical plants. In such uncertain environments, monitoring essential parameters like radiation, temperature, and gas levels is paramount, which can only be carried out autonomously via robot navigation. Visual saliency is now increasingly being utilized to assist robot navigation via an onboard camera, unlike earlier days when simultaneous localization and mapping (SLAM) was being carried out using the traditional Kalman and particle filter-based approaches. This chapter shows visual saliency determination applications utilizing cues like texture and color from acquired videos for robot navigation in uncertain environments. Although visual saliency determination for static camera cases has been studied earlier by estimating the motion map of moving objects, very few works exist on the role of global (camera) motion and its impact in determining the visual saliency for video sequences. Such a study investigating the effects of camera motion on visual attention can be performed; it would be of immense value for robotic applications as, generally, these robots move autonomously, monitoring some indoor/outdoor areas with cameras mounted on them. Since the camera dynamics in determining the visual saliency is less established, it opens up a new domain of research where investigations can be carried out to establish a relationship between the camera motion as a cue and the salient regions in a video sequence. In addition, a clear understanding of saliency determination in the presence of object and camera motion would be of great importance in solving many vision problems and valuable in many computer vision applications. One such application being targeted in this chapter is to utilize camera cues and determine visually salient regions so that a robot can navigate and monitor the essential parameters like radiation level or temperature level around a nuclear reactor or captive thermal power plants.