Smart Vision & Robotic Sensing

Professor, Robotics Laboratory
Smart Innovation Program, Graduate School of Advanced Science and Engineering
Hiroshima University
>> Research Contents
In order to establish high-speed robot senses that are much faster than human senses, we are conducting research and development of information systems and devices that can achieve real-time image processing at 1000 frames/s or greater. As well as integrated algorithms to accelerate sensor information processing, we are also studying new sensing methodologies based on vibration and flow dynamics; they are too fast for humans to sense.

High-Speed Real-Time 1000-Marker Extraction Algorithm

In this study, a labeling algorithm was implemented with high-speed hardware processing. The algorithm was actually installed as hardware on H3 Vision for verification.

For an object with several markers, the images of which can be binarized based on information such as intensity and color area. To extract of marker position information, a labeling moment calculation algorithm is installed on the high-speed vision platform H3 Vision for hardware processing.

Labeling and tracking at 1000fps are realized for an image of 1024×1024 pixels. And it at 10000fps are realized for an 256×256 pixels image. The algorithm has the following characteristics; (1) One scan algorithm: The 0-th and 1-st moments are calculated simultaneously at labeling, and Processing is completed by one scan. This makes it unnecessary to hold processing results or image size information as temporary information during calculation and allows processing on a little memory in hardware; (2) Reuse of label: Labels that are unnecessary after relabeling are recovered and reused when pixels that require labeling appear. This minimizes the memory area used for labeling; (3) Parallel processing: An image of 1024×1024 pixels is divided into 16 blocks of 1024×64 pixels, and a labeling moment calculation circuit is assigned to each block to increase the processing speed by parallel processing.

Color extraction processing, labeling moment calculation, and marker tracking were installed in the H3 Vision system to realize multipoint monochrome/color marker extraction and tracking. This consequently realized a high-speed real-time multipoint marker extraction system capable of extracting up to 4032 markers at 1000 fps and 254 markers at 10000 fps.

Extraction of 1000 white markers (1000fps) MPEG movie (6.5M)
Extraction of 1000 white markers (1000fps)
複数のカラーマーカー位置を抽出(1000fps) MPEG movie (7.7M)
Extraction of multiple color markers (1000fps)
Tracking of color markers on a fan rotation at 26rps (1000fps) MPEG movie (4.4M)
Tracking of color markers on a fan rotation at 26rps (1000fps)
Measurement of Guitar-String Vibration (10000fps) MPEG movie (3.8M)
Measurement of Guitar-String Vibration (10000fps)