Publications

The transportation system utilizes various methods to analyze passenger flow effectively. Although passenger information is accessible through the automated fare collection system, accurate live passenger counts at specific points require an onboard sensor system. The proposal emphasizes the necessity of a camera-based system, utilizing computer vision technology to count passenger. The paper also examines various computer vision methods, depth sensors, and tracking algorithms for implementing passenger counters in the transport sector. The proposed work introduced a wide-angle TOF sensor-based passenger tracking algorithm and obtained an accuracy of 98%. The method also explores a multi-tracking scenario after eliminating multiple sensors.

The passenger information system in railway sector is essential for helping passengers with schedules, connectivity, and overall comfort. It delivers on-board passenger information for trams, trains, and buses as an innovative solution with real-time information. Our proposed method utilizes an edge computing based computer vision algorithm to accurately count passenger data on an embedded computing board, including the number of passengers entering and exiting, as well as the current load in each carriage. This information is then presented in a way that helps passengers choose the carriage or route that best maintains their comfort. The system’s design and development, from the sensor device to the API specification and its visual representation, are developed as part of a comprehensive product development process. The proposed method tested with an accuracy of 98% using data collected from a wide angle Time-of-Flight (TOF) camera-based counting system that utilized passenger detection and tracking.

The goal of this research is to integrate an artificial intelligence framework for predicting Kathakali mudras, a crucial component of the traditional Indian dance style that is renowned for its complex hand and facial motions. Our goal is to protect and improve the accessibility of this traditional art form by utilising deep learning technology. To be more precise, we used the object identification model YOLOv8 together with convolutional neural networks (CNNs) like VGG 19, Inception V3, and ResNet 152 to predict and analyse Kathakali mudras. Furthermore, few-shot learning (FSL) and support vector machines (SVMs) were used to offer a comparative viewpoint on the efficacy of machine learning methods in this situation. The study entailed a thorough analysis of how well-performing different deep learning techniques performed. The predictive accuracy of VGG 19, Inception V3, and ResNet 152 for intricate mudras of Kathakali was subjected to a thorough evaluation process. In addition, an evaluation of YOLOv8’s real-time object identification capabilities was conducted in order to investigate its potential in live performance circumstances. The results of the experiment showed that VGG 19 performed better than the other models and had the highest accuracy in mudra prediction. As a result, VGG 19 was chosen as the best algorithm for this particular application. The results of this study demonstrate how artificial intelligence (AI) technologies may support and preserve traditional artistic forms. This AI-integrated framework is a useful tool for scholars, educators, and dancers, promoting a deeper knowledge and enjoyment of this cultural legacy through its accurate analysis and prediction of Kathakali mudras.

Sign language is a visual language that serves as the primary means of communication in the non-hearing community. It is essential to research and develop sign language translation methods to enable smooth communication between the non-hearing and the hearing communities. The non-hearing community faces several challenges that hinder their interaction with the general public, such as the lack of knowledge of sign language and the availability of interpreters. A Sign Language Converter would thus be an important tool in breaking the communication barrier between the non-hearing and hearing communities. The paper aims to build a model that converts Indian Sign Language into corresponding words. We plan on using a skeletal-point feature extraction framework to identify hand landmarks from sequences containing distinct signs and use these landmarks to build a model for recognizing hand gestures using various Long Short-Term Memory (LSTM) Networks. This approach can produce an accurate result compared to the traditional approach. The dataset is made by the co-authors of this paper due to the insufficiency of pre-existing ISL datasets. The user will be monitored and using the machine learning techniques discussed above, which will perform the real-time translation to display the final result.

The primary goal of this work is to develop a framework for short-term passenger flow prediction for metro rail transport systems. A reliable prediction of short-term passenger flow could greatly support metro authorities’ decision process. Both inflow and outflow of the metro stations are strongly associated with the travel demand within metro networks. Sequestered station-wise analysis ignores the spatial correlations existing between the stations. This paper tries to merge the spatial with the temporal by employing an indirect method of computing flow through O-D estimates for the same. Path-depended station-pairs of O-D flow are considered for employing a customized LSTM network. Experimental results indicate that the proposed passenger flow prediction model is capable of better generalization on short-term passenger flow than standard models of learning compared. This work also establishes that O-D prediction provides an indirect estimation procedure for passenger flow. The specific use case for this work is Kochi Metro Rail Limited (KMRL). A highlight of the work is that the whole analytics and modelling procedures are written on a customized scalable big-data platform (Jaison Paul Data Analytics Platform) JP-DAP which was developed prior to this work.

Transportation is an unavoidable part of every human’s life. The mobility system handles the transport of humans from different places using various transport modes. According to a station in a populated area, the main problem is the presence of traffic in peak hours and wasting their valuable time on the road. The only medium which runs above the traffic is metro rails/subways. For these reasons, metro rails become a point of interest for each researcher’s prophecy and provide valuable recommendations for the smooth functioning of services. Even though, in many cases, the metro systems are affected by abnormal passenger flow. So, this study handles abnormal passenger flow detection and station clustering for the behavior study of a passenger flow system. The research compares outlier detection and anomaly identification for the behavioral analysis of the metro rail passenger flow. The study use data from Kochi Metro Rail Limited for the period 2017 to 2019. Outlier removal has used in passenger flow data before building a forecasting system. In pattern recognition algorithm those components which lie outside the patterns can be considered abnormal (anomaly). The outliers are the component falling apart from the region of interest. The effect of removing the outlier from the time-series pattern is studied against the outlier included pattern to show the improvement.

In the field of transportation planning and management, passenger flow analysis is a significant problem with a wide range of applications. The prediction performance of forecast models is hence cardinal to any software analytic system. A predominant source of metro data is the automated fare card (AFC) system from which it is possible to gather a tremendous amount of information connected to passenger flow. Passenger flow represents a process whose dynamics are highly stochastic and dependent on a number of extrinsic and intrinsic parameters. This paper presents a restricted and simple model to study the intrinsic statistical influences governing the dynamics. These influences are either spatial or temporal. The feature space in which analysis algorithms run will be more effective if there is a collation of information from both spatial and temporal dimensions. The passenger flow parameter is fed into the layers of the deep neural network using the ST-LSTM (Spatio-Temporal Long Short-Term Memory) architecture. The architecture is evaluated with passenger movement data collected from the AFC information from the Kochi metro rail. To reduce the impact of irregular flow, the design uses the SVM-based outlier detection and elimination algorithm. A higher precision has been reached by the approach in comparison with SVR, ANN, LSTM algorithms.

This paper presents JP-DAP (Jaison-Paul Data Analytics Platform), an intelligent data analytics platform for metro rail transport systems designed to improve operations, customer experience, ridership forecasting, and administration by integrating and analysing multiple data sources.

JP-DAP includes a middleware built on HDFS and Apache Spark and uses open-source tools such as Apache Hive, Pandas, TensorFlow, and Spark MLlib for real-time and batch data processing. Benchmarking with TestDFSIO shows performance in line with industry standards.

A case study with Kochi Metro Rail Limited (KMRL) demonstrates analyses of Automated Fare Collection data, producing descriptive statistics and visualisations including inflow/outflow analysis, weekday/weekend travel patterns, and origin–destination matrices.

The platform supports short-term passenger flow prediction using Support Vector Regression (linear, RBF, and polynomial kernels); experiments indicate the linear kernel yields the most accurate next-day passenger count predictions using the previous five weekdays. Station one-to-all usage prediction using Long Short-Term Memory (LSTM) models is also integrated. Analytical results and visualisations are exposed via REST APIs and a web dashboard.

Metro rail systems are increasingly becoming relevant and inevitable in the context of rising demand for sustainable transportation methods. Metros are therefore going to have a consistently expanding user-base and hence user satisfaction will require meticulous planning. Usage forecast is clearly an integral component of metro planning as it enables forward looking and efficient allocation of resources leading to greater commuter satisfaction. An observation from studying the usage of Kochi Metro Rail Ltd. is that there is a consistently occurring temporal pattern in usage for every station. But the patterns differ from station to station. This hinders the search for a global model representing all stations. We propose a way to overcome this by using station memorizing Long Short-Term Memory (LSTM) which takes in stations in encoded form as input along with usage sequence of stations. This is observed to significantly improve the performance of the model. The proposed architecture with station parameter is compared with algorithms like SVR (support vector regression) and neural network implementation with the best architecture to testify the claim. The proposed model can predict the future flow with an error rate of 0.00127 MSE (mean squared error), which is better than the other models tested.

Liveness detection in mobile biometric is a challenging issue in iris recognition system security. RGB iris image is used for the image acquisition. In this work feature extracted between the genuine and fake by comparing the chromatic feature, blurred feature and pupil displacement. High quality printed iris images are considering the presentation attacks in this project. The printed images on glossy and matte paper, images shown in laptop, tablet screen with high resolution. Pupil localization technique using one dimensional processing of the eye region is evaluated. SVM classifier is used to classify the live or fake one.

Communication provide interaction among the people to exchange the feelings and ideas. The deaf community suffer a lot to interact with the community. Sign language is the way through which the people communicate with each other. In order to provide interaction with normal people there is a system which can convert the sign languages to the understandable form. The purpose of this work is to provide a real-time system which can convert Indian Sign Language (ISL) to the text. Most of the work based on handcrafted feature. In this we are introducing a deep learning approach which can classify the sign using the convolutional neural network. In the first phase we make a classifier model using the numeral signs using the Keras implementation of convolutional neural network using python. In phase two another real-time system which used skin segmentation to find the Region of Interest in the frame which shows the bounding box. The segmented region is feed to the classifier model to predict the sign. The system has attained an accuracy of 99.56% for the same subject and 97.26% in the low light condition. The classifier found to be improving with different background and the angle of the image captured. Our method focus on the RGB camera system.

Human Activity Recognition has a lot of applications such as patient monitoring, rehabilitation and assisting disabled. When mobile sensors are held to the subject’s body, they permit continuous monitoring of numerous signal patterns from the phone. This has appealing use in healthcare applications. In order to improve the state of global healthcare, numerous healthcare devices have been introduced that allow doctors to perform remote monitoring and increase users’ motivation and awareness. Nowadays smart phones have become a part of our day to day life. The best way to implement this idea is through smart phones. The smart phones contain various built-in sensing units like accelerometer, gyroscope, GPS, compass sensor and barometer. Using this, a system is designed to capture the states of a user. Here the mobile sensor is used as an input device and we estimate the human motion activity using data mining and machine learning techniques. We use the KNN classification algorithm in the activity recognition system which supports the training and classification using accelerometer data only. We can predict the performance of these classifiers from a series of observations on human activities like walking, running, step up, and step down in an activity recognition system.