Indrajeet Ghosh
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Indrajeet Ghosh has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Nirmalya Roy
Information Systems, University of Maryland Baltimore County, Baltimore, MD, USA
8 shared publications - Garvit Chugh
Indian Institute of Technology Jodhpur, Jodhpur, India
3 shared publications - Sreenivasan Ramasamy Ramamurthy
Department of Computer Science, Bowie State University, USA
3 shared publications Avijoy Chakma
Mobile Pervasive & Sensor Computing Lab, Department of Information Systems University of Maryland Baltimore County (UMBC) Baltimore Maryland 21250 USA
2 shared publications- Kasthuri Jayarajah
Computer Science, New Jersey Institute of Technology, Newark, NJ, USA
2 shared publications - Mohammad Saeid Anwar
Department of Information Systems , USA
2 shared publications
Latest Publications
- Open Access
- 17 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Autonomous and Adaptive Systems
MobHeteroCAS: Mobility-Aware DNN Task Scheduling in Heterogeneous Collaborative Autonomous Systems
Scene perception for situational awareness is crucial for autonomous agents like UGVs and UAVs that involves sensing from different 2D and 3D modalities to execute tasks such as semantic segmentation and object detection. However, UGVs and UAVs face limitations in power and computation capacity. To address these challenges, we propose an offloading framework called MobHeteroCAS , designed to leverage nearby peer autonomous vehicles for computation. Our framework addresses key challenges in offloading DNN (Deep Neural Networks) tasks in a collaborative environment: (a) Optimally offloading data/features to leverage nearby devices’ computation capabilities; (b) Reducing offloaded data for efficient communication in low-bandwidth environments; (c) MobHeteroCAS determines the optimal ”split ratio” to determine how much data should be offloaded to each nearby device based on network parameters such as RSSI values, latency incurred due to network conditions, workload, memory constraints (CPU, GPU, RAM), and power limitations. The key contribution of MobHeteroCAS is determining the optimal split ratio in stationary and mobile environments. Apart from the devices consuming more power and memory when they are mobile, the latency increases when the distance between communicating devices increases. Thus, MobHeteroCAS estimate the latency to offload the computation when devices are on mobile before selecting the optimal split ratio. MobHeteroCAS is truly fault-tolerant, incorporating the identification of device (agent) failure and task redistribution when a device fails. We empirically demonstrate that offloading DNN features instead of raw images reduces overall system latency. Therefore, we offload feature representations of raw images for tasks such as image classification and semantic segmentation. In our study, we utilize a testbed comprising three Unmanned Ground Vehicles (UGVs) and one Linux laptop, implementing a self-adaptive optimization framework capable of managing multiple heterogeneous tasks collaboratively across multiple nodes. Our evaluations, involving simultaneous execution of multiple DNN tasks like Image Classification and Semantic Segmentation, demonstrate that employing a split ratio of 30%-42%-28% (data distribution across auxiliary nodes) reduces offloading latency by approximately 72% (from 38 \(ms/data\) to 10.5 \(ms/data\) ) compared to compressed frames and reduces total operation time by approximately 99% (from 281.33 minutes to 169 seconds) compared to the baseline configuration (execution on the primary node) in stationary setting. The latency improves by 78-82%.- 17 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Autonomous and Adaptive Systems
Open AccessMobHeteroCAS: Mobility-Aware DNN Task Scheduling in Heterogeneous Collaborative Autonomous Systems
Scene perception for situational awareness is crucial for autonomous agents like UGVs and UAVs that involves sensing from different 2D and 3D modalities to execute tasks such as semantic segmentation and object detection. However, UGVs and UAVs face limitations in power and computation capacity. To address these challenges, we propose an offloading framework called MobHeteroCAS , designed to leverage nearby peer autonomous vehicles for computation. Our framework addresses key challenges in offloading DNN (Deep Neural Networks) tasks in a collaborative environment: (a) Optimally offloading data/features to leverage nearby devices’ computation capabilities; (b) Reducing offloaded data for efficient communication in low-bandwidth environments; (c) MobHeteroCAS determines the optimal ”split ratio” to determine how much data should be offloaded to each nearby device based on network parameters such as RSSI values, latency incurred due to network conditions, workload, memory constraints (CPU, GPU, RAM), and power limitations. The key contribution of MobHeteroCAS is determining the optimal split ratio in stationary and mobile environments. Apart from the devices consuming more power and memory when they are mobile, the latency increases when the distance between communicating devices increases. Thus, MobHeteroCAS estimate the latency to offload the computation when devices are on mobile before selecting the optimal split ratio. MobHeteroCAS is truly fault-tolerant, incorporating the identification of device (agent) failure and task redistribution when a device fails. We empirically demonstrate that offloading DNN features instead of raw images reduces overall system latency. Therefore, we offload feature representations of raw images for tasks such as image classification and semantic segmentation. In our study, we utilize a testbed comprising three Unmanned Ground Vehicles (UGVs) and one Linux laptop, implementing a self-adaptive optimization framework capable of managing multiple heterogeneous tasks collaboratively across multiple nodes. Our evaluations, involving simultaneous execution of multiple DNN tasks like Image Classification and Semantic Segmentation, demonstrate that employing a split ratio of 30%-42%-28% (data distribution across auxiliary nodes) reduces offloading latency by approximately 72% (from 38 \(ms/data\) to 10.5 \(ms/data\) ) compared to compressed frames and reduces total operation time by approximately 99% (from 281.33 minutes to 169 seconds) compared to the baseline configuration (execution on the primary node) in stationary setting. The latency improves by 78-82%. - Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-19
MorsEar: Toward Generalizable Low-Resource Covert Messaging via Earable based Inertial Sensing
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-19
Open AccessMorsEar: Toward Generalizable Low-Resource Covert Messaging via Earable based Inertial Sensing
- Open Access
- 19 January 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
SkillNet: Human Actions Assessment via Human-AI Collaboration
Intelligent human motion analysis is essential for developing next-generation IoT and AR/VR systems that enable automated, interpretable, and fine-grained performance assessment. Motivated by the need for real-time, explainable, and transferable skill evaluation, we propose a wearable sensing framework to assess human performance by tracking skill progression and minimizing injury risk. We use live badminton gameplay and workout exercises as representative use cases, where motion dynamics, postural stability, and limb coordination are critical to success. Both activities demand optimal posture and synchronized limb movements, while improper actions or suboptimal technique can lead to decreased performance and higher injury susceptibility. We introduce SkillNet , a multi-task learning framework that extracts shared representations across all limbs while preserving limb-specific motion signatures. The architecture employs task-specific regressors to detect subtle inter-limb dissimilarities and distinctive traits, enabling collective inference in a body sensor network (BSN) environment. To holistically measure performance, we formulated a weighted performance indicator (PI) that fuses AI-driven scoring with domain-expert evaluations, providing a robust metric for both qualitative and quantitative assessment. We evaluate SkillNet on three diverse datasets B adminton A ctivity R ecognition (BAR), M ulti- M odalities D ataset of S ports (MMDOS), and D aily and S ports A ctivities (DSADS) capturing a broad spectrum of motion types and skill intensities. Results show that SkillNet achieves an R \({}^{2}\) score of 86% and a mean squared error of 0.0093 in performance prediction. The integrated AI–expert scoring mechanism improves baseline performance estimation by 14.95% , demonstrating the advantage of combining human expertise with automated analysis. We further benchmark inference time, memory usage, and power consumption of the SkillNet , validating its efficiency and feasibility for real-time, end-to-end task inference on resource-constrained embedded edge devices, Jetson Nano and Jetson Xavier NX platforms.- 19 January 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
Open AccessSkillNet: Human Actions Assessment via Human-AI Collaboration
Intelligent human motion analysis is essential for developing next-generation IoT and AR/VR systems that enable automated, interpretable, and fine-grained performance assessment. Motivated by the need for real-time, explainable, and transferable skill evaluation, we propose a wearable sensing framework to assess human performance by tracking skill progression and minimizing injury risk. We use live badminton gameplay and workout exercises as representative use cases, where motion dynamics, postural stability, and limb coordination are critical to success. Both activities demand optimal posture and synchronized limb movements, while improper actions or suboptimal technique can lead to decreased performance and higher injury susceptibility. We introduce SkillNet , a multi-task learning framework that extracts shared representations across all limbs while preserving limb-specific motion signatures. The architecture employs task-specific regressors to detect subtle inter-limb dissimilarities and distinctive traits, enabling collective inference in a body sensor network (BSN) environment. To holistically measure performance, we formulated a weighted performance indicator (PI) that fuses AI-driven scoring with domain-expert evaluations, providing a robust metric for both qualitative and quantitative assessment. We evaluate SkillNet on three diverse datasets B adminton A ctivity R ecognition (BAR), M ulti- M odalities D ataset of S ports (MMDOS), and D aily and S ports A ctivities (DSADS) capturing a broad spectrum of motion types and skill intensities. Results show that SkillNet achieves an R \({}^{2}\) score of 86% and a mean squared error of 0.0093 in performance prediction. The integrated AI–expert scoring mechanism improves baseline performance estimation by 14.95% , demonstrating the advantage of combining human expertise with automated analysis. We further benchmark inference time, memory usage, and power consumption of the SkillNet , validating its efficiency and feasibility for real-time, end-to-end task inference on resource-constrained embedded edge devices, Jetson Nano and Jetson Xavier NX platforms. - Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 718-728
High-Order Moments Conditional Domain Adaptation Networks for Wearable Human Activity Recognition
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 718-728
Open AccessHigh-Order Moments Conditional Domain Adaptation Networks for Wearable Human Activity Recognition
- Open Access
- 13 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 11-21
Augmenting Personalized Memory via Practical Multimodal Wearable Sensing in Visual Search and Wayfinding Navigation
- 13 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 11-21
Open AccessAugmenting Personalized Memory via Practical Multimodal Wearable Sensing in Visual Search and Wayfinding Navigation
- Open Access
- 18 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 142-151
β-Decode: Attention-based Decoding Temporal Artifacts via Unsupervised β-Variational Autoencoder
- 18 December 2024
- Published by Association for Computing Machinery (ACM)
- p. 142-151
Open Accessβ-Decode: Attention-based Decoding Temporal Artifacts via Unsupervised β-Variational Autoencoder
- Top Cited Papers
- 21 March 2023
- Published by Wiley in WIREs Data Mining and Knowledge Discovery
- Vol. 13 (5)
Sports analytics review: Artificial intelligence applications, emerging technologies, and algorithmic perspective
The rapid and impromptu interest in the coupling of machine learning (ML) algorithms with wearable and contactless sensors aimed at tackling real‐world problems warrants a pedagogical study to understand all the aspects of this research direction. Considering this aspect, this survey aims to review the state‐of‐the‐art literature on ML algorithms, methodologies, and hypotheses adopted to solve the research problems and challenges in the domain of sports. First, we categorize this study into three main research fields: sensors, computer vision, and wireless and mobile‐based applications. Then, for each of these fields, we thoroughly analyze the systems that are deployable for real‐time sports analytics. Next, we meticulously discuss the learning algorithms (e.g., statistical learning, deep learning, reinforcement learning) that power those deployable systems while also comparing and contrasting the benefits of those learning methodologies. Finally, we highlight the possible future open‐research opportunities and emerging technologies that could contribute to the domain of sports analytics.This article is categorized under: Technologies > Machine Learning Technologies > Artificial Intelligence Technologies > Internet of Things
- 21 March 2023
- Published by Wiley in WIREs Data Mining and Knowledge Discovery
- Vol. 13 (5)
Top Cited PapersSports analytics review: Artificial intelligence applications, emerging technologies, and algorithmic perspective
The rapid and impromptu interest in the coupling of machine learning (ML) algorithms with wearable and contactless sensors aimed at tackling real‐world problems warrants a pedagogical study to understand all the aspects of this research direction. Considering this aspect, this survey aims to review the state‐of‐the‐art literature on ML algorithms, methodologies, and hypotheses adopted to solve the research problems and challenges in the domain of sports. First, we categorize this study into three main research fields: sensors, computer vision, and wireless and mobile‐based applications. Then, for each of these fields, we thoroughly analyze the systems that are deployable for real‐time sports analytics. Next, we meticulously discuss the learning algorithms (e.g., statistical learning, deep learning, reinforcement learning) that power those deployable systems while also comparing and contrasting the benefits of those learning methodologies. Finally, we highlight the possible future open‐research opportunities and emerging technologies that could contribute to the domain of sports analytics.This article is categorized under: Technologies > Machine Learning Technologies > Artificial Intelligence Technologies > Internet of Things
- 14 May 2022
- Published by Elsevier in Pervasive and Mobile Computing
- Vol. 83
DeCoach: Deep Learning-based Coaching for Badminton Player Assessment
- 14 May 2022
- Published by Elsevier in Pervasive and Mobile Computing
- Vol. 83
DeCoach: Deep Learning-based Coaching for Badminton Player Assessment
- Open Access
- 1 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
StanceScorer: A Data Driven Approach to Score Badminton Player
In recent times, wearable devices have gained immense popularity for IoT applications, especially for sports analytics. Recent works in sports analytics primarily focuses on improving a player's performance and help devise a winning strategy based on the player's strengths and weaknesses which is also the objective of this paper. In a racquet-based sports, it is often assumed that handling the racquet majorly influences the performance of the players, however, the stance and the posture of the player are of greater importance. A perfect posture and stance allow a player to play a stroke efficiently by directing the shuttle to strategic spots. Therefore, it helps to utilize less energy and make it difficult for the opponent to return the shot and eventually score a point. Hence, we hypothesize that the performance of a player equally correlates with the stance and the efficiency of handling the racquet. In this paper, we propose to analyze the stance of the player based on the shot played. In an attempt to do so, we propose a data-driven approach to evaluate a player's performance based on the player's stance or posture. First, we employ both shallow learning and deep learning algorithms to classify the strokes which is then used to analyse the stance. Secondly, we propose a distance based methodology to compare the stance of an intermediate or a novice player with that of a professional player. Further, we learn the error between the professional player's stance with that of a participant and propose a scoring methodology. To evaluate our proposed methodology, we deploy a sensor network comprising of inertial measurement units (IMU) sensors on the dominant wrist and palm; and both the legs. We collect the data from a professional player, an intermediate player and a novice player for 12 different frequently played shots and evaluate our proposed methodology with this dataset.
- 1 March 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-6
Open AccessStanceScorer: A Data Driven Approach to Score Badminton Player
In recent times, wearable devices have gained immense popularity for IoT applications, especially for sports analytics. Recent works in sports analytics primarily focuses on improving a player's performance and help devise a winning strategy based on the player's strengths and weaknesses which is also the objective of this paper. In a racquet-based sports, it is often assumed that handling the racquet majorly influences the performance of the players, however, the stance and the posture of the player are of greater importance. A perfect posture and stance allow a player to play a stroke efficiently by directing the shuttle to strategic spots. Therefore, it helps to utilize less energy and make it difficult for the opponent to return the shot and eventually score a point. Hence, we hypothesize that the performance of a player equally correlates with the stance and the efficiency of handling the racquet. In this paper, we propose to analyze the stance of the player based on the shot played. In an attempt to do so, we propose a data-driven approach to evaluate a player's performance based on the player's stance or posture. First, we employ both shallow learning and deep learning algorithms to classify the strokes which is then used to analyse the stance. Secondly, we propose a distance based methodology to compare the stance of an intermediate or a novice player with that of a professional player. Further, we learn the error between the professional player's stance with that of a participant and propose a scoring methodology. To evaluate our proposed methodology, we deploy a sensor network comprising of inertial measurement units (IMU) sensors on the dominant wrist and palm; and both the legs. We collect the data from a professional player, an intermediate player and a novice player for 12 different frequently played shots and evaluate our proposed methodology with this dataset.