Visual grid refinement in targeting useful data in moving visual data for mobile robots

This paper discuss the problem face during extracting necessary information from a 640 × 480 visual grid coming from a single forward facing camera. Latency and excessive processing of the visual grid limits the frame rate which robot reacts to incoming obstacles and line tracing. By refining the data grid to limit only significant locations where movement is most active during the robots exploration of the environment, quicker response can be achieved. The refined data grid is then used as an input pipeline to supply information to a neural network that will generate autonomous behaviors.