Md. Shoaib Bhuiyan

Estimating heart rate using wrist-type Photoplethysmography and acceleration sensor while running

This study provides Heart Rate (HR) Estimation using wrist-type Photoplethysmogpraphy (PPG) sensor while the subject is running. We propose the algorithm to estimate heart rate for the wrist-type PPG sensor. Since body motion artifacts easily affect the arm portion, our method in this study also uses accelerometer built in the wrist-type sensor to improve the accuracy of heart rate estimation. Our method has two components. One is rejecting artifacts with the power spectrums difference between PPG and acceleration obtained by frequency analysis. The other is the reliability of heart rate estimation, defined by the acceleration. Experimental results while our test subjects were running came closer to the holter Electrocardiogram (ECG) in high accuracy (r = 0.98, SD = 8.7 bpm). We, therefore, report the heart rate estimation method which has a higher degree of usability compared to existing methods using ECG.

Hand alphabet recognition using morphological PCA and neural networks

Proposes a method of feature extraction based upon the principal component analysis (PCA) of the pixel positions for the description of the hand postures from colored glove images. We analyze its performance applying it to a neural network based Japanese and American manual alphabet recognition system, while the background remains natural. Average recognition rates of 89.1% for the Japanese and 93.3% for the American fingerspelling has been obtained for a set of 42 Japanese kana and 26 international hand alphabet postures respectively, using a feedforward multilayer perceptron neural neural classifier.

Estimation of drowsiness level based on eyelid closure and heart rate variability

This paper presents a novel method that uses eyelid closure and heart rate variability to estimate the drivers drowsiness level. Laboratory experiments were conducted by using a proprietary driving simulator, which induced drowsiness among the test drivers. The purposes of these experiments were to obtain the electrocardiogram (ECG) and the eye-blink video sequences. Also the drivers were monitored through a video camera. The changes in facial expression of the drivers were used as a standard index of drowsiness level. Error-Correcting Output Coding (ECOC) was employed as a multi-class classifier to estimate the drowsiness level. We extended the ordinary ECOC using a loss function for decoding procedure to obtain class tendencies of each drowsiness level. We used the Loss-based Decoding ECOC (LD-ECOC) to classify the drowsiness level. As a result, we obtained an extraordinarily high accuracy for estimation of drowsiness level.

Optimal Edge Detection under Difficult Imaging Conditions

This paper incrementally extends the energy minimization techniques for image analysis developed by [Koch et al. 1986]. Our application is edge extraction and we use the dual intensity and line processes introduced by [Geman and Geman, 1984]. The approach seeks to minimize a global energy functional that explicitly incorporates image properties to be minimized into weighted terms of the energy functional. Our specific contribution is modifying the weighting of terms in the energy functional that were previously independent of spatial gray level change to explicitly include spatial change in the weighting. We argue that the weighting used in previous implementations resulted in a reduced contribution from the edge components due to a dominance of the spatial intensity difference term as that spatial difference increases in size. Our specific modification compensates for this effect by scaling the edge process weighting factors by the spatial difference value (to the second order), thus, maintaining the same relative effect as the spatial difference increases. We found that the proposed algorithm works significantly better as compared to Koch et al. because of this modification.

Driver Assistance Systems to Rate Drowsiness: A Preliminary Study

This paper attempts to present a comprehensive survey of what is being done to automate the drowsiness ratings to be employed within a vehicle. The paper analyses the evidences for the usefulness of the measures currently used in drowsiness detection devices, which are not invasive and is based solely on eye activity. Their relationships with drowsiness and performance are described, and general problems and pitfalls associated with their practical use in passenger vehicles are identified. It also simulates a non-intrusive drowsiness detection system that is the core detection technique of several devices under review to understand how all the components of the system respond in real-time. A rating table to aid in automating the drowsiness rating in future is also included based upon analysis of drowsiness observed from recorded video.

Edge detection by neural network with line process

Though edge detection is an indispensable part in image processing, no definite method still exists. Existing methods can not detect edge precisely when contrast changes largely within the object due to non-uniform illumination. Geman and Geman applied a line process to express the discontinuity in gray level of images. Variation of contrast in an image still remains a problem. Koch et al. developed an energy equation whose coefficients remained constant. This paper presents an edge detection method for images where contrast is not uniform. It shows an way of changes in coefficients of the line process energy equation. Besides, use of neural network helps to reduce noise, thereby making human intervention unnecessary in the detection of edge.

Estimation of drivers' drowsiness level using a Neural Network Based ‘Error Correcting Output Coding’ method

This paper presents an improved method for estimating a drivers drowsiness level using eyelid closure and heart rate variability. Laboratory experiments were conducted by using a proprietary driving simulator, which induced drowsiness among the test drivers. The purposes of these experiments were to obtain an electrocardiogram (ECG) and eye-blink video sequences. The drivers were also monitored through a video camera. The changes in facial expression of the drivers were used as a standard index of drowsiness level. Error-Correcting Output Coding (ECOC) was employed as a multi-class classifier to estimate drowsiness level. The ordinary ECOC method was extended using a loss function for the decoding procedure to obtain class tendencies of each drowsiness level. We used a Neural Network-Based Decoding method to classify the drowsiness level. As a result, we obtained an extraordinarily high accuracy for estimation of drowsiness level. Comparisons were also made with the result of those from using the Loss-based Decoding ECOC (LD-ECOC).

T-CombNET - A Neural Network Dedicated to Hand Gesture Recognition

T-CombNET neural network structure has obtained very good results in hand gesture recognition. However one of the most important setting is to define an input space that can optimize the global performance of this structure. In this paper the Interclass Distance Measurement criterion is analyzed and applied to select the space division in T-CombNET structure. The obtained results show that the use of the IDM criterion can improve the classification capability of the network when compared with practical approaches. Simulations using Japanese finger spelling has been done. The recognition rate has improved from 91.2% to 96.5% for dynamic hand motion recognition.

Hand gesture recognition using morphological principal component analysis and an improved CombNET-II

A new neural network structure dedicated to time series recognition, T-CombNET, is presented. The model is developed from a large scale neural network CombNet-II, designed to deal with a very large vocabulary for character recognition. Our specific modifications of the original CombNet-II model allows it to do temporal analysis, and to be used in a large set of human movement recognition systems. This paper also presents a feature extraction method based on morphological principal component analysis that completely describes a hand gesture in 2-dimensional time varying vector. The proposed feature extraction method along with the T-CombNET structure were then used to develop a complete Japanese Kana hand alphabet recognition system consisting of 42 static postures and 34 hand motions. We obtained a superior recognition rate of 99.4% in the gesture recognition experiments when compared to different neural network structures like multi-layer perceptron, learning vector quantization (LVQ), Elman and Jordan partially recurrent neural networks, CombNET-II and the proposed T-CombNET structure.

Estimation of a three-dimensional gaze point and the gaze target from the road images

Various ways of measuring a glance and estimating a gaze point have been studied in the literature, but they are mainly of two-dimension. We propose a new method to estimate three-dimensional information of a gaze target and to obtain the appearance of the gaze target. We propose the use of two sets of stereo cameras. One is used to measure the three-dimensional environment outside the vehicle, and the other to measure the drivers glance. By integrating these stereo camera information, we have estimated the three-dimensional coordinates of the drivers gaze. We also obtain information such as a forward vehicle, a sign, pedestrian and so on of the gaze target from the projection into a road image.