Computer Vision @ LEMS

This paper describes a new median filter algorithm optimized for real-time performance in smart cameras with embedded processors. As in the JPEG and MPEG compression algorithms, each frame of the video stream is first partitioned into a regular array of non-overlapping square blocks. The median value for each block is then computed and compared with corresponding values of neighboring blocks. If the magnitude of the difference does not exceed a threshold, the output value for all the pixels in the block is set to the median value. Otherwise, the output value for each pixel in the block is computed as the median value within a window of the same size centered at this pixel. We describe variations for binary and grayscale images. The algorithm has been implemented and tested in an embedded single-board-computer (SBC) with no hardware acceleration, as a component of a Visual Sensor Network (VSN) system for real-time indoor person detection and tracking. In this system, where the SBCs have the additional overhead of decoding JPEG frames from IP cameras, our new algorithm is 5 to 20 times faster than the traditional algorithms for typical window sizes. We expect further speedups to frame-rate performance on smart cameras with embedded image sensors and reconfigurable hardware.