Ryosuke Konishi

This paper describes an intelligent wheelchair with a novel interface which uses visual oral motion. The wheelchair runs in various lighting environments, traditional mouth region detection method did not consider these conditions. To achieve detection of mouth region and analysis mouth motion in real-time and in various lighting environments, we proposed mouth cavity region detection method. In our system, the user moves his mouth open and close to run the wheelchair. We carried out six operational experiments and running experiments. Experiments with five subjects obtained our system is supposed to use a certain extent in a campus corridor.

The visual information doesn't have an influence with the noise in comparison with the auditory information. However, lip-reading is earlier technology than speech recognition, there is a problem that recognition rate is low. This paper proposes a novel feature to obtain high recognition rate for lip-reading. The proposed lip-reading method is as follows. First, sampled active contour model is applied to detect lip contour during the input image sequences. The lip size is then normalized. Next, a time change of two lip shape features (area and aspect ratio) is expressed as a two-dimensional trajectory of the lip motion. The most similar trajectory in a database which is compared with the target trajectory by DP matching is selected as a result word. We recorded ten Japanese words that uses for the wheelchair control. As a result of comparative experiment with other features in which has already been proposed, trajectory feature obtained the highest recognition rate of 97.2%

This paper describes an automatic detection and tracking method for unknown object from unclear background. Our method consists of two processes, the detection process and the tracking process. The former process is the background subtraction method and the block-based clustering. The latter process is with mean shift method. To detect an accurate size of the target region, we set a correction period. Demonstration was carried out with a scene which three men are appeared and walked. It is found that two regions could be tracked even when two men was walked toward different direction which they walking together. We confirmed that the proposed method could process on real time.

Recently, speech, especially word recognition using visual information, has attracted significantly interest. In word recognition, the target is not just a word, but also a vowel. However, since vowel frames do not contain many phonemes, vowel recognition rate is less than that of the word. Some research has been done on vowel recognition. This paper analyses features that can effectively recognize five Japanese vowels. The process of our method is as follows. First, the sampled active contour model is applied to detect the lip contour during the input image sequences. The lip size is then normalized. Next, various features, such as the shape, diameter, and the approximate function features, are calculated and compared with the recognition rates using k nearest neighbor method. Two experiments, speaker dependent recognition and speaker independent recognition are carried out with five subjects. We calculated eleven feature sets and found that the feature set including the area and aspect ratio is the most effective. In the speaker dependent recognition, we obtained the recognition rate of 96.6%, and in the speaker independent recognition, we obtained 82.4%

This paper describes the lip reading method using video and thermal images. We pay attention to the lip movement. The density of the thermal image expresses the surface temperature, and it is difficult to detect the lip position automatically. Though the video and the thermal images take in the same time, these don't synchronized in time and spatially. Then the time matching and the spatial matching process are applied. Next, a lip region is detected from the video image, and a lip region of the thermal image is detected corresponding to the video region. Finally, the eigenimage waveform is computed, and DP matching is applied. Experiments were conducted for 75 scenes (five words of three subjects). A successful recognition rate of 76.0% with the video image, 44.0% with the thermal image, and 80.0% with both images were obtained. Experimental results show the recognition rate was improves by combining with the video and the thermal images

This paper describes the development of the monocular autonomy following vehicle which for reduce a helper burden. The system is composed by a CCD camera, a FPGA board and some control circuits. Our idea is that the distance between the system and the target human is related to the size of the human region in the camera image. Based on this idea, we proposed the automatic human region detection method with S-ACM which is a kind of active contour model. We implemented our method and developed the system. The experiment which followed the target human was done. As the result, we confirmed that the system followed with keeping a fixed distance with the target human

The volatile organic compound (VOC) is contained in indoor materials in buildings, such as paints and adhesives. The formaldehyde is a representative example of the chemical substances which affect people’s health. For detection of formaldehyde, we made SnO₂ thin-film type sensors by the DC magnetron sputtering method. The sensor has Pt electrodes on one side of a glass substrate as the heater. The thickness of the heater is 500 nm. The SnO₂ thin film is sputtered on the opposite side of the heater. The thickness of SnO₂ thin film is 200 nm. Ag(1.6 wt.%) is added as a catalyst metal. As formaldehyde is low concentration of less than 1 ppm, the output signal from sensor is buried in noise, which is generated from the heater. The noise is considered as white noise. In this paper, noise cancellation method by using Adaptive Line Enhancer(ALE) is proposed to extract the sensor outputs. The least-square method (LMS) is adopted for the adaptive algorithm in the ALE. As a result of noise cancellation by using the ALE, the formaldehyde of 0.1 ppm concentration can be detected.

A gas discrimination system which can estimate the concentration of the discriminated gases are proposed in this paper. TMS320C50 DSP by Texas Instruments and semiconductor gas sensor TGS823 by FIGARO Electronic are utilized for a data processing unit and a sensing probe, respectively. As for software, an artificial neural network and a fuzzy-neural network are used for the discrimination and the concentrationestimation of introduced gases, respectively. Hydrogen, vapor of alcohol and cigarette smoke are selected as domestic gas examples in the experiment, and the efficiency of the proposed system is confirmed according to the experimental results.

Word recognition techniques are investigated and proposed by several researchers. According to these techniques, relatively high recognition rates can be achieved in calm circumstances. But in noisy circumstances, it becomes difficult to accomplish high recognition rate. In this paper, we develop a real time lip reading system on a commercial personal computer to provide word recognition method in the very noisy circumstances. To realize a real time lip reading system, very fast extraction of lip shapes from face image sequence is indispensable. In the proposed system, the active contour model with vibration factor is introduced and is modified to solve this problem. Automatic extraction method of word speaking period and a novel lip shape recognition parameter are also proposed. In the proposed system, HMMs are used for recognition of lip shape movements. To confirm the effectiveness of proposed system, recognition experiments of 10 Japanese railroad station names are accomplished. The experimental results are included.