Cloud computing and visual attention based object detection for power substation surveillance robots

Visual object detection is an important issue of outdoor autonomous robots. The detection results can be used for recognizing an object, tracking a specified object and construction a map of environments, etc. In the substation surveillance application, the detection results can be used to locate the target object and segment out from the whole image for next analysis. Visual attention, which has drawn much attention of researchers in the machine vision field, is an effective tool for object detection. It tends to reduce the search area by simulating the mechanism of human vision which can focus on the significant target quickly and others are ignored. However, to compute the visual attention saliency map is a resource consumption work, the onboard hardware capacity limits the whole system performance. Additionally, hundreds of surveillance robots are working in different substation, a continually updated target library needs to be shared with each robot, it is impossible for the onboard resources to maintain all the historical data and learning from it. In this paper, a cloud based visual attention objects detection method was proposed. Cloud computing technology is adopted to extend the computing and storage ability of the local robots. All the data can save in the cloud and the computing resources can be on demand dynamically allocated. Furthermore, a visual attention based object detection algorithm is also proposed, to obtain the most salient region (MSR) in the whole image, then the SURF matching is implemented to detect the object, if the target object is inside MSR, it will be segmented out not only for analysis, but also used for update the target library. Finally, the implementation results are provided.