Gaze tracking is important to measure what attracts people's interest. However, the person being tracked needs to wear the measurement instrument, or to stay limited area with the conventional gaze tracking systems. This limitation is expected to be eliminated if the telescopic image of one's eye can be continuously tracked and observed. In this presentation, we propose an algorithm for tracking human eyes at high speed with 1 ms auto pan tilt system, and report the experiment result.
In the existing rust detection method using machine learning, training data was constructed by human vision judgment. Therefore, the objectivity of the training data was insufficient, which would reduce the accuracy of the detection result. In this study, more objective rust training data is constructed using the light reflection characteristics of rust and iron plate. The rust detection performance by a typical machine learning method is evaluated using the constructed training data. In addition, rust detection experiments were carried out with proposed method, and its effectiveness was demonstrated.
A questionnaire survey was conducted about the effect of blue motifs on white plates, including the number of motifs and the presence of a blue rim, on appetite by placing Japanese food on small serving plates. Results showed that blue rim motifs stimulated the appetite better, and two motifs were often evaluated better than a single motif. A blue rim and motifs may bring the colors of the dish closer together and balance the arrangement and coloration of foods. There was a tendency for lower evaluations when the motifs were hidden by the dish; therefore, it is desirable to place dishes so that the motifs are visible. Furthermore, the optimum blue proportions that make the dish look better are different between patterns of a blue plate and motifs on a white plate: about 40％ and 16-20％ of the entire plate, respectively. An additional questionnaire revealed that when dishes with patterns and motifs were compared side by side, dishes with motifs went better with Japanese dishes served on small plates.
An image processing method focusing saturation to recognize achromatic surgical tools such as forceps and energy devices in an endoscopic image has been developed. The proposed method could be used for semi-automatic control of a laparoscopic robot. It makes use of an extracting information for saturation from the HLS image, difference between channels of original RGB image, a masking process based on binarization, closing, removal of the boundary, labeling, and recognition of the point in the surgical tool. The point recognized in three laparoscopic surgical videos using this method were performed, and the recognition rate was 50 % for four conditions of surgical tools.
A part of the surface mount work of printed circuit boards is conducted by human workers. Therefore, when adopting new employees, it is required to conduct a training for practicing the work to them. The present training methods is the OJT with instruction by a manager or a expert worker. That method requires much personnel cost. The goal of this study is to develop an instruct system for the surface mount work of printed circuit boards by utilizing a CCD camera, image processing techniques, a laser irradiation device and so on, which can conduct a training in place of a manager or a expert worker. This paper describes the design of the instruction system, developed functions and the results of their evaluations.