Disc degeneration is a common type of lumbar disc disease. Disc degeneration leads to low back pa... more Disc degeneration is a common type of lumbar disc disease. Disc degeneration leads to low back pain, and it is caused due to injury in Intervertebral Disc (IVD). An automatic diagnostic system to diagnose degenerative discs from T2-weighted sagittal MR image is proposed. A fully automated Expectation-Maximization (EM)-based new IVD segmentation is proposed to segment the lumbar IVD from mid-sagittal MR image. Then, a hybrid of basic intensity, invariant moments, Gabor features are extracted from segmented IVDs. The IVDs are classified as degenerative or non-degenerative using Support Vector Machine (SVM) classifier. The proposed system is trained, tested and evaluated for 93 clinical sagittal MR images of 93 patients. The optimized hyperparameters are estimated. The proposed model is tested and validated for the dataset and obtained an accuracy of 92.47%. The patient-based analysis was performed and obtained an accuracy of 92.86%. The performance analysis of the proposed model with other classifiers like k-NN, decision tree, Linear Discriminant Analysis (LDA) and Feedforward neural network is also analyzed. This proposed method outperforms when compared with state-of-the-art methods. This system can be used as a second opinion in diagnosing degenerative discs.
2016 International Conference on Inventive Computation Technologies (ICICT), 2016
Emergency exit signs are used for safety purpose in buildings, malls, shops, etc and various othe... more Emergency exit signs are used for safety purpose in buildings, malls, shops, etc and various other places. They provide escape routes or ways using which people can escape during emergency situations. Escaping during such situations using the emergency exit signs and emergency doors are easier for sighted people, but visually impaired should rely upon someone else. Hence, in this paper we propose an approach to recognize the emergency exit sign using a mobile phone. This involves developing software that recognizes the exit sign captured by the mobile camera and intimidate the direction to escape as an audio output. This software is packaged as an android application on a mobile phone making it easily accessible and less expensive for the visually impaired people. In our approach, we have used edge detection and region detection techniques to identify possible object candidates in the complex image captured. We, then identify exit sign board by checking the presence of the word using ‘EXIT’ using region filtering, which filters out non-text and non-arrow regions and optical character recognition. By analyzing the connected components present in the exit sign board region, we separate the text and arrow region. Using template matching, we find the direction of the arrow, which is then rendered as an audio output.
Proceedings of the International Conference on ISMAC in Computational Vision and Bio-Engineering 2018 (ISMAC-CVB), 2019
Formation of abnormal cells in brain serves the major cause of tumor. With estimated deaths of 22... more Formation of abnormal cells in brain serves the major cause of tumor. With estimated deaths of 229,000 as of 2015, it has become an issue to be dealt. The less awareness of brain tumor owes to lots of unaccounted deaths. Thus, we aim in developing an app which could serve the purpose of detecting the tumor and giving additional information related to the detected tumor. This app takes in an MRI image and does preprocessing followed by clustering, segmentation, and binarization. The preprocessing involves the conversion of the image into grayscale and noise filtering. We aim at using improved Fuzzy c-means algorithm for clustering and segmentation. Binarization mainly aims at calculating the tumor size useful for further analysis. The improved fuzzy c-means algorithm overcomes the various constraints of k-means algorithm such as time complexity, processing of noisy images, and memory space.
Change detection is the art of quantifying the changes in Synthetic Aperture Radar (SAR) images o... more Change detection is the art of quantifying the changes in Synthetic Aperture Radar (SAR) images occurring over a period of time. Remote sensing has been instrumental in performing change detection analysis. The impact of applying the combination of texture features for classification techniques to separate water bodies from land masses is empirically investigated in this paper. First, the images are classified using unsupervised Principle Component Analysis (PCA) based K-means clustering for dimension reduction. Then, texture features like Energy, Entropy, Contrast, Inverse Difference Moment, Directional Moment and Median are extracted using Gray Level Co-occurrence Matrix (GLCM) and these features are utilized in Learning Vector Quantization (LVQ) and Support Vector Machine (SVM) classifiers. This paper aims to apply a combination of the texture features in order to significantly improve the accuracy of detection. The utility of detection analysis influences management and policy d...
Categorization of water bodies and land areas from the satellite image is performed since the pre... more Categorization of water bodies and land areas from the satellite image is performed since the prediction of satellite image has become a major challenging issue due to weather condition, atmosphere, etc. Previously, data mining is used for clustering in various application such as text data, similarities in images and bioinformatics data. In this paper, a novel approach has been designed by incorporating the PSO and DE algorithm for data mining technique in the satellite image. Here feature extraction is carried out by using DWT, PCA, and GLCM techniques. In the proposed method, an optimized PSO-DE algorithm is designed to obtain the best solution in order to get the better satellite data. Finally, the estimated output is compared with the existing method on the bases of performances, and it is found to be efficient. The performance parameters such as PSNR, MSE, RMS, mean, variance, correlation, contrast, energy, homogeneity, SD, and entropy are evaluated for the Landsat and MODIS satellite images.
One of the common types of lumbar disc disease is bulging which cause low back pain, tingling, an... more One of the common types of lumbar disc disease is bulging which cause low back pain, tingling, and numbness. An automatic diagnostic system to detect the disc pathology would be helpful to the radiologist. A computer aided diagnostic system is proposed to identify the disc bulge in axial lumbar spine MR images. A new EM based segmentation method is applied to segment the Intervertebral Disc (IVD) from the axial slice of T2-weighted MRI. After segmentation, the features are extracted by executing Histogram of Oriented Gradients (HOG) and a novel feature descriptor called as Intervertebral disc Descriptor (IdD). The features obtained are trained by Support Vector Machine(SVM). In this work, T2-weighted axial slices of lumbar MR images for 93 patients are used for evaluation. The proposed framework is trained, tested and validated on 675 clinical axial MR images of 93 patients, in which 184 are normal, 55 are herniated, and 436 are bulged images. On applying the proposed system, an accuracy of 92.78% is obtained for classifying normal and bulge and compared with different classifiers such as k-nn, decision trees and feed forward neural network. This model produces high accuracy, sensitivity, specificity, and fscore to detect bulge in the MRI. The model built with SVM produces a better result when compared with k-nn, decision trees and feed forward neural network. Also, the same model can be applied to detect other disc pathologies such as desiccation and degeneration.
International Journal of Information Technology, 2019
Sonar, the instrument used for acquiring underwater acoustic images strongly participate in the a... more Sonar, the instrument used for acquiring underwater acoustic images strongly participate in the assistance of detection and recognition of objects under the seafloor. Sonar emits sound waves to navigate deep into the sea and detects the sunken objects. The sonar images are also used for fish habitat mapping. Noise is an important factor that contributes to the degradation of quality of the images obtained by sonar. Generally speckle noise is found in the acoustic images which are caused by the instruments that affects the quality, thereby reducing visual perception. In this paper, various spatial filtering techniques have been applied to the acoustic images to remove the speckle noise. Among the filtering techniques available, bilateral filter followed by guided filter, when applied to the acoustic images tend to remove the speckle noise to a greater degree. The statistical methods for image quality assessment such as mean squared error (MSE), peak signal to noise ratio (PSNR), Structural SIMilarity Index (SSIM) are used to compare the quality of the despeckled images.
Multidimensional Systems and Signal Processing, 2017
Dredging the surface of the ocean to identify both living and non living things nowadays has beco... more Dredging the surface of the ocean to identify both living and non living things nowadays has become an unproblematic task with the help of the acoustic instruments. Side scan sonar is one of such instruments used for far-reaching the seafloor. The sonar captures the scene of the sea bed by releasing fan shaped sound signal which is then converted to images. These images are normally gray scale low contrast images where the objects cannot be viewed clearly. The proposed method uses the Stationary Wavelet Transform (SWT) to decompose the input image into four components such as Low-Low, Low-High, High-Low and High-High components. The low frequency component is sharpened using Laplacian filter and a mask is created by subtracting the LL component with the filtered image. Then the enhanced LL component is obtained by adding the mask to the input image. The high contrast image is reconstructed by applying inverse stationary wavelet transform which combines the enhanced LL component and the other sub-bands. The results have been compared by replacing the SWT with the Discrete Wavelet Transform by interpolating the frequency components. The quantitative and visual results show that the proposed method using SWT outperforms the state of art techniques in terms of contrast.
Proceedings of the Second International Conference on Information and Communication Technology for Competitive Strategies, 2016
Segmentation techniques have received great attention for object detection in image processing an... more Segmentation techniques have received great attention for object detection in image processing and it uses the edge detection as a vital step. Edge detection methods aim at identifying points in an image at which the image brightness changes sharply. There are recent methods that are employed for detecting the edges in digital images which do not suit for low resolution underwater acoustic images. For detecting the edges in the acoustic images, a method which uses wiener filtering followed by median filtering for smoothing and morphological image processing techniques is proposed. The morphological dilation finds the local maximum value and morphological erosion determines the local minimum value in an image. The difference between the dilated image and eroded image gives the morphological gradient. The final edge map is obtained by applying the binarization method to the gradient image. The resultant image has pixel values with increased contrast intensity in the close neighborhood. The edge map shows better performance with respect to the results of the conventional edge detection techniques such as Sobel and Canny.
2015 International Conference on Applied and Theoretical Computing and Communication Technology (iCATccT), 2015
Image enhancement techniques play a vital role in improving the quality of the low resolution ima... more Image enhancement techniques play a vital role in improving the quality of the low resolution images. There are recent methods that are employed for enhancing the digital images which do not suit for low resolution underwater acoustic images. Many common approaches used in vision and pattern recognition turned to be inefficient because of the nature of the noise present in acoustic images. For enhancing the acoustic images, a method which uses discrete wavelet transform to decompose the image into Low-Low, Low-High, High-Low and High-High components is proposed. Then the Low-Low frequency component is enhanced using Karhumen-Loeve (K-L) transform to estimate a new enhanced LL component. The difference image is obtained by subtracting the enhanced LL component with the original LL component. Other three high frequency components are also interpolated and added with the difference image. These enhanced components are reconstructed using Inverse discrete wavelet transform. The resultant enhanced image has better contrast and resolution compared to the images obtained by Generalized histogram equalization (GHE) and by using Singular Value Decomposition (SVD) approach. The experimental results show that the proposed system outperforms the GHE and SVD methods in terms of image enhancement.
2019 2nd International Conference on Power and Embedded Drive Control (ICPEDC)
This paper presents the recognition of hand gesture in the research field of machine vision. Visi... more This paper presents the recognition of hand gesture in the research field of machine vision. Vision based hand gesture recognition has the capacity to develop a tool for Human Machine Interaction (HCI). The automated threshold methods were used as pre-processing steps for extraction of feature vector using chain code histogram (CCH). Then, construct the kernel based on histogram of chain code using density measure to obtain discriminative feature descriptor for efficient recognition of hand gesture using Support Vector Machine (SVM). Cluster based threshold techniques involves Otsu thtresholding (OT), Ridler and Calvard thresholding (RCT), and Kittler and Illingworth thresholding (KIT) are used to segment the region of interest for feature extraction. In this paper, CCH based on various segmentation methods were compared to measure the recognition rate by SVM classifier. The proposed RCT-CCH based kernel method increase the recognition rate of hand posture by 90%, compared with cluster based thresholds.
2021 Sixth International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET)
Pedestrian detection and driver assistance applications draws an important role on safety measure... more Pedestrian detection and driver assistance applications draws an important role on safety measure such as speed control, crash control, sensing crash and occupant location detection. The pedestrian accidents occurs due to the vulnerable traffic users such as humans, stranded or moving vehicle or other obstacles. To avoid the pedestrian accident, this paper proposes a model that can accomplish pedestrian detection automatically using Histogram of Gradient (HOG) and You Only Look Once (YOLO) algorithm. The experiments are carried out on Forward Looking Infrared Radar (FLIR) starter thermal dataset consisting of 5000 images. The HOG algorithm is implemented on these thermal image samples and is classified using Support Vector Machine (SVM) classifier. The accuracy of YOLO is calculated using intersection over union method between the ground truth and the predicted bounding box. To further improve the safety of the user an alarm is designed to alert the user on sight of pedestrians during night.
2018 International Conference on Computer, Communication, and Signal Processing (ICCCSP), 2018
This paper presents a recognition of sign language in the area of computer vision and pattern rec... more This paper presents a recognition of sign language in the area of computer vision and pattern recognition system. The local features of invariant images were extracted using speeded up robust features (SURF) with dimensionality reduction techniques. Then, K-nearest neighbour classification technique is used for establishing the recognition system. The local feature descriptor of SURF was computationally complex for classifying the word signs. Laplacian eigenmaps has been combined with SURF to reduce dimensionality of feature descriptor and computation time for classification. In this paper, execution of recognition rate has been developed by using Laplacian eigenmaps of dimensionality reduction compared with other methods of principal component analysis and singular value decomposition. By applying Laplacian eigenmaps, sign classification accuracy was improved from 90 to 96% than the dimensionality reduction strategy.
2017 IEEE International Conference on Power, Control, Signals and Instrumentation Engineering (ICPCSI), 2017
This paper presents a recognition system for understanding the American Sign Language using Suppo... more This paper presents a recognition system for understanding the American Sign Language using Support Vector Machine (SVM) and Error Correcting Output Codes (ECOC). In the pre-processing stage, invariant local features were extracted to develop the recognition system using multi-class support vector machine. The binary classifier of SVM employed with ECOC framework to deal multi-class problems. In ECOC, the encoding matrix was designed using various coding strategy such as one versus one, one versus all and random matrix. The data has trained based on coding matrix with set of dichotomizes independently and classify the data using loss-based decoding method to obtain efficient accuracy. In this paper, performance of recognition rate has been improved by updating given ECOC matrix from the original matrix using beam ECOC optimization. By employing beam ECOC, classification accuracy was increased from 70 to 80% than the state-of-art coding strategy.
Disc degeneration is a common type of lumbar disc disease. Disc degeneration leads to low back pa... more Disc degeneration is a common type of lumbar disc disease. Disc degeneration leads to low back pain, and it is caused due to injury in Intervertebral Disc (IVD). An automatic diagnostic system to diagnose degenerative discs from T2-weighted sagittal MR image is proposed. A fully automated Expectation-Maximization (EM)-based new IVD segmentation is proposed to segment the lumbar IVD from mid-sagittal MR image. Then, a hybrid of basic intensity, invariant moments, Gabor features are extracted from segmented IVDs. The IVDs are classified as degenerative or non-degenerative using Support Vector Machine (SVM) classifier. The proposed system is trained, tested and evaluated for 93 clinical sagittal MR images of 93 patients. The optimized hyperparameters are estimated. The proposed model is tested and validated for the dataset and obtained an accuracy of 92.47%. The patient-based analysis was performed and obtained an accuracy of 92.86%. The performance analysis of the proposed model with other classifiers like k-NN, decision tree, Linear Discriminant Analysis (LDA) and Feedforward neural network is also analyzed. This proposed method outperforms when compared with state-of-the-art methods. This system can be used as a second opinion in diagnosing degenerative discs.
2016 International Conference on Inventive Computation Technologies (ICICT), 2016
Emergency exit signs are used for safety purpose in buildings, malls, shops, etc and various othe... more Emergency exit signs are used for safety purpose in buildings, malls, shops, etc and various other places. They provide escape routes or ways using which people can escape during emergency situations. Escaping during such situations using the emergency exit signs and emergency doors are easier for sighted people, but visually impaired should rely upon someone else. Hence, in this paper we propose an approach to recognize the emergency exit sign using a mobile phone. This involves developing software that recognizes the exit sign captured by the mobile camera and intimidate the direction to escape as an audio output. This software is packaged as an android application on a mobile phone making it easily accessible and less expensive for the visually impaired people. In our approach, we have used edge detection and region detection techniques to identify possible object candidates in the complex image captured. We, then identify exit sign board by checking the presence of the word using ‘EXIT’ using region filtering, which filters out non-text and non-arrow regions and optical character recognition. By analyzing the connected components present in the exit sign board region, we separate the text and arrow region. Using template matching, we find the direction of the arrow, which is then rendered as an audio output.
Proceedings of the International Conference on ISMAC in Computational Vision and Bio-Engineering 2018 (ISMAC-CVB), 2019
Formation of abnormal cells in brain serves the major cause of tumor. With estimated deaths of 22... more Formation of abnormal cells in brain serves the major cause of tumor. With estimated deaths of 229,000 as of 2015, it has become an issue to be dealt. The less awareness of brain tumor owes to lots of unaccounted deaths. Thus, we aim in developing an app which could serve the purpose of detecting the tumor and giving additional information related to the detected tumor. This app takes in an MRI image and does preprocessing followed by clustering, segmentation, and binarization. The preprocessing involves the conversion of the image into grayscale and noise filtering. We aim at using improved Fuzzy c-means algorithm for clustering and segmentation. Binarization mainly aims at calculating the tumor size useful for further analysis. The improved fuzzy c-means algorithm overcomes the various constraints of k-means algorithm such as time complexity, processing of noisy images, and memory space.
Change detection is the art of quantifying the changes in Synthetic Aperture Radar (SAR) images o... more Change detection is the art of quantifying the changes in Synthetic Aperture Radar (SAR) images occurring over a period of time. Remote sensing has been instrumental in performing change detection analysis. The impact of applying the combination of texture features for classification techniques to separate water bodies from land masses is empirically investigated in this paper. First, the images are classified using unsupervised Principle Component Analysis (PCA) based K-means clustering for dimension reduction. Then, texture features like Energy, Entropy, Contrast, Inverse Difference Moment, Directional Moment and Median are extracted using Gray Level Co-occurrence Matrix (GLCM) and these features are utilized in Learning Vector Quantization (LVQ) and Support Vector Machine (SVM) classifiers. This paper aims to apply a combination of the texture features in order to significantly improve the accuracy of detection. The utility of detection analysis influences management and policy d...
Categorization of water bodies and land areas from the satellite image is performed since the pre... more Categorization of water bodies and land areas from the satellite image is performed since the prediction of satellite image has become a major challenging issue due to weather condition, atmosphere, etc. Previously, data mining is used for clustering in various application such as text data, similarities in images and bioinformatics data. In this paper, a novel approach has been designed by incorporating the PSO and DE algorithm for data mining technique in the satellite image. Here feature extraction is carried out by using DWT, PCA, and GLCM techniques. In the proposed method, an optimized PSO-DE algorithm is designed to obtain the best solution in order to get the better satellite data. Finally, the estimated output is compared with the existing method on the bases of performances, and it is found to be efficient. The performance parameters such as PSNR, MSE, RMS, mean, variance, correlation, contrast, energy, homogeneity, SD, and entropy are evaluated for the Landsat and MODIS satellite images.
One of the common types of lumbar disc disease is bulging which cause low back pain, tingling, an... more One of the common types of lumbar disc disease is bulging which cause low back pain, tingling, and numbness. An automatic diagnostic system to detect the disc pathology would be helpful to the radiologist. A computer aided diagnostic system is proposed to identify the disc bulge in axial lumbar spine MR images. A new EM based segmentation method is applied to segment the Intervertebral Disc (IVD) from the axial slice of T2-weighted MRI. After segmentation, the features are extracted by executing Histogram of Oriented Gradients (HOG) and a novel feature descriptor called as Intervertebral disc Descriptor (IdD). The features obtained are trained by Support Vector Machine(SVM). In this work, T2-weighted axial slices of lumbar MR images for 93 patients are used for evaluation. The proposed framework is trained, tested and validated on 675 clinical axial MR images of 93 patients, in which 184 are normal, 55 are herniated, and 436 are bulged images. On applying the proposed system, an accuracy of 92.78% is obtained for classifying normal and bulge and compared with different classifiers such as k-nn, decision trees and feed forward neural network. This model produces high accuracy, sensitivity, specificity, and fscore to detect bulge in the MRI. The model built with SVM produces a better result when compared with k-nn, decision trees and feed forward neural network. Also, the same model can be applied to detect other disc pathologies such as desiccation and degeneration.
International Journal of Information Technology, 2019
Sonar, the instrument used for acquiring underwater acoustic images strongly participate in the a... more Sonar, the instrument used for acquiring underwater acoustic images strongly participate in the assistance of detection and recognition of objects under the seafloor. Sonar emits sound waves to navigate deep into the sea and detects the sunken objects. The sonar images are also used for fish habitat mapping. Noise is an important factor that contributes to the degradation of quality of the images obtained by sonar. Generally speckle noise is found in the acoustic images which are caused by the instruments that affects the quality, thereby reducing visual perception. In this paper, various spatial filtering techniques have been applied to the acoustic images to remove the speckle noise. Among the filtering techniques available, bilateral filter followed by guided filter, when applied to the acoustic images tend to remove the speckle noise to a greater degree. The statistical methods for image quality assessment such as mean squared error (MSE), peak signal to noise ratio (PSNR), Structural SIMilarity Index (SSIM) are used to compare the quality of the despeckled images.
Multidimensional Systems and Signal Processing, 2017
Dredging the surface of the ocean to identify both living and non living things nowadays has beco... more Dredging the surface of the ocean to identify both living and non living things nowadays has become an unproblematic task with the help of the acoustic instruments. Side scan sonar is one of such instruments used for far-reaching the seafloor. The sonar captures the scene of the sea bed by releasing fan shaped sound signal which is then converted to images. These images are normally gray scale low contrast images where the objects cannot be viewed clearly. The proposed method uses the Stationary Wavelet Transform (SWT) to decompose the input image into four components such as Low-Low, Low-High, High-Low and High-High components. The low frequency component is sharpened using Laplacian filter and a mask is created by subtracting the LL component with the filtered image. Then the enhanced LL component is obtained by adding the mask to the input image. The high contrast image is reconstructed by applying inverse stationary wavelet transform which combines the enhanced LL component and the other sub-bands. The results have been compared by replacing the SWT with the Discrete Wavelet Transform by interpolating the frequency components. The quantitative and visual results show that the proposed method using SWT outperforms the state of art techniques in terms of contrast.
Proceedings of the Second International Conference on Information and Communication Technology for Competitive Strategies, 2016
Segmentation techniques have received great attention for object detection in image processing an... more Segmentation techniques have received great attention for object detection in image processing and it uses the edge detection as a vital step. Edge detection methods aim at identifying points in an image at which the image brightness changes sharply. There are recent methods that are employed for detecting the edges in digital images which do not suit for low resolution underwater acoustic images. For detecting the edges in the acoustic images, a method which uses wiener filtering followed by median filtering for smoothing and morphological image processing techniques is proposed. The morphological dilation finds the local maximum value and morphological erosion determines the local minimum value in an image. The difference between the dilated image and eroded image gives the morphological gradient. The final edge map is obtained by applying the binarization method to the gradient image. The resultant image has pixel values with increased contrast intensity in the close neighborhood. The edge map shows better performance with respect to the results of the conventional edge detection techniques such as Sobel and Canny.
2015 International Conference on Applied and Theoretical Computing and Communication Technology (iCATccT), 2015
Image enhancement techniques play a vital role in improving the quality of the low resolution ima... more Image enhancement techniques play a vital role in improving the quality of the low resolution images. There are recent methods that are employed for enhancing the digital images which do not suit for low resolution underwater acoustic images. Many common approaches used in vision and pattern recognition turned to be inefficient because of the nature of the noise present in acoustic images. For enhancing the acoustic images, a method which uses discrete wavelet transform to decompose the image into Low-Low, Low-High, High-Low and High-High components is proposed. Then the Low-Low frequency component is enhanced using Karhumen-Loeve (K-L) transform to estimate a new enhanced LL component. The difference image is obtained by subtracting the enhanced LL component with the original LL component. Other three high frequency components are also interpolated and added with the difference image. These enhanced components are reconstructed using Inverse discrete wavelet transform. The resultant enhanced image has better contrast and resolution compared to the images obtained by Generalized histogram equalization (GHE) and by using Singular Value Decomposition (SVD) approach. The experimental results show that the proposed system outperforms the GHE and SVD methods in terms of image enhancement.
2019 2nd International Conference on Power and Embedded Drive Control (ICPEDC)
This paper presents the recognition of hand gesture in the research field of machine vision. Visi... more This paper presents the recognition of hand gesture in the research field of machine vision. Vision based hand gesture recognition has the capacity to develop a tool for Human Machine Interaction (HCI). The automated threshold methods were used as pre-processing steps for extraction of feature vector using chain code histogram (CCH). Then, construct the kernel based on histogram of chain code using density measure to obtain discriminative feature descriptor for efficient recognition of hand gesture using Support Vector Machine (SVM). Cluster based threshold techniques involves Otsu thtresholding (OT), Ridler and Calvard thresholding (RCT), and Kittler and Illingworth thresholding (KIT) are used to segment the region of interest for feature extraction. In this paper, CCH based on various segmentation methods were compared to measure the recognition rate by SVM classifier. The proposed RCT-CCH based kernel method increase the recognition rate of hand posture by 90%, compared with cluster based thresholds.
2021 Sixth International Conference on Wireless Communications, Signal Processing and Networking (WiSPNET)
Pedestrian detection and driver assistance applications draws an important role on safety measure... more Pedestrian detection and driver assistance applications draws an important role on safety measure such as speed control, crash control, sensing crash and occupant location detection. The pedestrian accidents occurs due to the vulnerable traffic users such as humans, stranded or moving vehicle or other obstacles. To avoid the pedestrian accident, this paper proposes a model that can accomplish pedestrian detection automatically using Histogram of Gradient (HOG) and You Only Look Once (YOLO) algorithm. The experiments are carried out on Forward Looking Infrared Radar (FLIR) starter thermal dataset consisting of 5000 images. The HOG algorithm is implemented on these thermal image samples and is classified using Support Vector Machine (SVM) classifier. The accuracy of YOLO is calculated using intersection over union method between the ground truth and the predicted bounding box. To further improve the safety of the user an alarm is designed to alert the user on sight of pedestrians during night.
2018 International Conference on Computer, Communication, and Signal Processing (ICCCSP), 2018
This paper presents a recognition of sign language in the area of computer vision and pattern rec... more This paper presents a recognition of sign language in the area of computer vision and pattern recognition system. The local features of invariant images were extracted using speeded up robust features (SURF) with dimensionality reduction techniques. Then, K-nearest neighbour classification technique is used for establishing the recognition system. The local feature descriptor of SURF was computationally complex for classifying the word signs. Laplacian eigenmaps has been combined with SURF to reduce dimensionality of feature descriptor and computation time for classification. In this paper, execution of recognition rate has been developed by using Laplacian eigenmaps of dimensionality reduction compared with other methods of principal component analysis and singular value decomposition. By applying Laplacian eigenmaps, sign classification accuracy was improved from 90 to 96% than the dimensionality reduction strategy.
2017 IEEE International Conference on Power, Control, Signals and Instrumentation Engineering (ICPCSI), 2017
This paper presents a recognition system for understanding the American Sign Language using Suppo... more This paper presents a recognition system for understanding the American Sign Language using Support Vector Machine (SVM) and Error Correcting Output Codes (ECOC). In the pre-processing stage, invariant local features were extracted to develop the recognition system using multi-class support vector machine. The binary classifier of SVM employed with ECOC framework to deal multi-class problems. In ECOC, the encoding matrix was designed using various coding strategy such as one versus one, one versus all and random matrix. The data has trained based on coding matrix with set of dichotomizes independently and classify the data using loss-based decoding method to obtain efficient accuracy. In this paper, performance of recognition rate has been improved by updating given ECOC matrix from the original matrix using beam ECOC optimization. By employing beam ECOC, classification accuracy was increased from 70 to 80% than the state-of-art coding strategy.
Uploads
Papers by Sree Sharmila