Real-Time Line Detection Method for Camera Images of the Rescue Robot

Abstract

Scene analysis based on the images captured from rescue robot is used to be composed of the basic three image processing stages: edge detection, line detection, and object recognition. The key technology to do this is Hough transform for line detection. The FIHT3 method, a companion method of a Fast Incremental Hough Transform (FIHT), is very appropriate to such line detection, because it is the fastest in processing and is with highly accurate in performance. However, the processing time allowed by line detection is 10 ms or less to analyze the scene in real time. Because the FIHT3 method has achieved speed-up by the shift operation, the number of partitions in θ axis of Hough voting space can take only a certain limited value. Therefore, the FIHT3 method should be more improved so that it could be applicable to any number of partitions in θ axis. Because FIHT3 is not applicable to the odd number of angle partitions due to the basic property of the simultaneous recurrence formula. Then, in this research, we propose the improved FIHT method that can process even if the number of angle partitions is odd. Furthermore, in the experiments of rescue robot it was known experimentally that the improved FIHT method provides the complete accuracy and the fastest computation among conventional any other methods.