Gait Silhouette Enhancement with Modified CLAHE and Precise Gait Recognition Using a Lightweight Convolutional Neural Network International Journal of Intelligent Systems and Applications in Engineering, 2024
Gait speed based individual recognition model using deep 2-D convolutional neural network Abhay Lal, P Nithyakani 2023 International Conference on Computer Communication and Informatics Iccci 2023, 2023 GAIT represents the pattern in which a person walks. We have four types of walking conditions: normal walking, slow walking, fast walking, and normal walking with a bag. GAIT analysis has been beneficial to doctors to analyse the patterns of walking to detect pains and diseases. In this work, we will compare the classification of the four types of conditions of GAIT using 2-D Convolutional neural networks. The CASIA C dataset has been used to train the 2-D Convolutional neural networks that have images captured using thermal infrared night imaging. Using transfer learning, we tested the performance of various CNN algorithms like DenseNet201 and MobileNetV2. The various algorithms provide us with information that the GAIT C dataset is best for the custom CNN model created and we have compared all the results in this paper.
Age and Gender Prediction using Adaptive Gamma Correction and Convolutional Neural Network Anweasha Saha, S Nithish Kumar, P Nithyakani 2023 International Conference on Computer Communication and Informatics Iccci 2023, 2023 The classification of age and gender has drawn increased attention recently because of its significance in creating user-friendly intelligent systems. In the domains of image processing and computer vision, determining age from a single facial image has proven a challenging job. Convolutional Neural Network (CNN) based techniques have been frequently adopted for the classification problem in the recent past because of their precise results in facial analysis. This study incorporates an end-to-end CNN approach with the addition of a key pre-processing step for image contrast enhancement, which was done via an adaptive gamma correction technique to produce precise gender and age group classification of real-world faces. The complete feature extraction and classification processes are included in the two-level CNN architecture. The feature extraction task pulls features that are related to gender and age while the classification assigns the facial photographs to the proper gender and age group. The proposed network has been trained and tested on the Adience (original) dataset. The results of the experiment seem to back up the claim that the proposed model is better at classifying people by gender and age when the Adience benchmark for classification is used.
Multimodal Deep Neural Network for Diabetic Retinopathy Detection Vijay Ravichander, Aditi Mittal, P Nithyakani 2023 International Conference on Computer Communication and Informatics Iccci 2023, 2023 An eye condition known as diabetic retinopathy (DR) destroys the retinal blood vessels and impairs vision. Normal, mild, moderate, severe, and PDR are the five stages of DR (Proliferative Diabetic Retinopathy). The risk of significant vision loss can be reduced by nearly 90% with early diagnosis and treatment of DR. A highly skilled practitioner typically diagnoses DR by looking at the retina’s vivid fundus pictures. Diabetic Retinopathy and its various stages can be identified automatically utilizing snapshots of the retina with various computer-vision-based techniques. These techniques however cannot classify the early phases of DR. This study employs a hybrid deep learning architecture using Inceptionv3 and Inception – ResNetv2 to diagnose all five phases of DR while maintaining high accuracy levels for the early stages. The proposed system consists of two steps. First off, it anticipates the early phases of DR (normal, mild or higher). The technique will further detect for stage 3, 4, and 5, which are moderate, severe, and PDR, if the third stage is first anticipated (Proliferative Diabetic Retinopathy). The initial and final stages of DR may both benefit from this methodology’s ability to maintain strong performance.
Human Gait recognition Using Cross View Micro Gait Dataset with Light weight MobileNet Haripreeth Dwarakanath Avarur, Abhay Lal, Nithyakani P, Aryan Sinha, Gajulapalli Naga Vyshnavi, Shruthi Kannan 2023 International Conference on Recent Advances in Electrical Electronics Ubiquitous Communication and Computational Intelligence Raeeucci 2023, 2023 Gait Analysis in human identification is a pivotal biometric feature which has recently drawn attention in the modern world. Currently, surveillance cameras (in Airports, Banks, etc.) do not always capture the front-view of a human. To resolve the current issue, gait analysis is used to recognize a person. In this study, machine learning and deep learning model are utilized to recognize the human with their gait. Cross View Micro Gait (CVM-GAIT) Dataset is created with numerous individual recorded videos in the cross view with various speeds, which has been converted into frames and stored as images. This study is carried out with SVM, Decision tree, Inception net and the proposed Lightweight Mobile net architecture. The results prove that the proposed model outperforms the state of art with live recorded video.
Classification Of Gait Pathology Using Enhanced Convolutional Neural Network Nithyakani P, Ferni Ukrit M 2022 International Conference on Computer Communication and Informatics Iccci 2022, 2022 Human gait analysis has become a manifold approach to identify gait impairments. Normal human gait has the ability to walk or move without any difficulty. Abnormal human gait maximize the effort to maintain the stability to walk. This abnormality is caused by numerous pathological factors. Gait analysis depend on the expertise to identify the abnormality. Deep learning approach has become more popular and successful approach in image classification, prediction and so on. In order to expedite the challenging gait analysis, deep learning algorithms is utilized in this research work. Enhanced convolutional neural network is proposed and built to identify the gait pathologies with less number of parameters in comparison with pre-trained CNN model. Sagittal view of gait sequence is taken as an input to the proposed model. The layers of the proposed model were optimized to increase the performance. The proposed method is tested over INIT gait dataset, DAI gait dataset and DAI gait dataset 2. The experimental results exhibit that the enhanced convolutional neural network is reliable to classify the gait sequence into pathology in effective manner. The proposed model outperforms 97.4 % accuracy in classifying the gait pathologies of DAI gait 2 dataset when compared with INIT and DAI gait dataset.
Music Genre Recognition Using Short Time Fourier Tranform And CNN Tanmay Toshniwal, Parisha Tandon, Nithyakani P 2022 International Conference on Computer Communication and Informatics Iccci 2022, 2022 Music can unite networks and rise above language obstructions. Anyone can access it at their fingertips, whether through internet or musical apps. Rising interest in automatic music genre classification originates from budding demand for systematical organization of audio files from abundant digital music available online. Music genre classification and playlist generator has the major function of detecting and grouping music of similar genre. Music theory accepts several rules that help human order, for example, harmonies and musical structures, instrumental game plan and so on. Be that as it may, in general, melodic substance is intricate and music kinds are not all around characterized, making it a difficult AI issue, and identify it from various other gathering issues. The target of this paper is to improve the presentation of old style calculations, including Logistic Regression (LR), Gaussian Discriminant Analysis (GDA), Random Forest (RF) what's more, Support Vector Machine (SVM), by consolidating them with a fine tuned VGG-16 convolutional neural network. We preprocess the raw music signal with Short Time Fourier Transform (STFT) to obtain the fine-tuned spectrogram for classification. We obtained 88.5% of accuracy and evaluated the loss using categorical cross entropy.
Prediction of Bitcoin Price Using Bi-LSTM Network Nithyakani P, Rijo Jackson Tom, Piyush Gupta, A Shanthini, Vivia Mary John, Vipul Sharma 2021 International Conference on Computer Communication and Informatics Iccci 2021, 2021 Machine Learning and Artificial Intelligence based money exchanging have pulled in enthusiasm in the recent years with the introduction of Bitcoins. The cost of Bitcoins has increased in a large scale and it is fairly difficult to predict the future cost per Bitcoin. In this study, we utilize a machine learning and deep learning model to analyze the digital currency market to predict the cost of Bitcoin per day. We dissect everyday information for 1,691 cryptographic forms of money for the period between November 2017 and April 2019. The study shows that straightforward exchanging procedures assisted by best in class AI algorithms have met the standard benchmarks. Our outcomes also show that non-inconsequential, basic algorithmic instruments can help in envision of momentary development of the cryptographic money. The proposed system uses a Bi- directional LSTM for forecasting the bitcoin prices. The proposed model was able to trace the test dataset with Mean Absolute Percentage Error of 13%. The model is helpful for the user to take decision on investing in Bitcoins.
