Muralikrishnan Mani

Muralikrishnan Mani

Auto-generated profile, based on publicly available information

Muralikrishnan Mani

Abhira

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 July 2025 in IEEE Access
Verified
Journal Article
Open access
SRSonaa Rajagopal
Sonaa Rajagopal
MMMuralikrishnan Mani
Muralikrishnan Mani
SVShyam Venkatraman
Shyam Venkatraman
RSR. Suganya
R. Suganya
Foot diseases such as plantar fasciitis and flat feet occur in millions of people worldwide, resulting in mobility problems and serious health complications. This paper presents an AI-based approach that uses machine learning to identify foot ailments, classify foot conditions, and suggest the right footwear. The approach integrates clinical gait analysis, symptom diagnosis, and customized footwear suggestions with synthetically generated images, which aid in the production of the footwear. The proposed work is based on three large datasets: (1) Grayscale pressure sensor heat maps for foot posture, with high-resolution foot pressure maps that capture weight distribution and posture; (2) Clinically Validated Foot Condition Dataset, comprising foot conditions verified by physiotherapists and linked to real symptoms; and (3) Footwear Recommendation Dataset for Specific Foot Conditions, with expert-curated footwear suggestions tailored to various foot conditions for optimal support and comfort. The framework consists of three modules: a CNN-based VGG-16-GRU model which identifies gait posture based on pressure sensor heatmaps, an autoencoder-based random forest model which classifies foot diseases based on the detected gait posture, and an LSTM-ensembled XGBoost model which recommends features of suggested footwear. The recommended footwear features obtained from this model is then synthetically generated to visually perceive the suggested footwear. The experimental results show excellent performance with detecting and recommending shoe designs. Additionally, Stable Diffusion-based synthetically generated footwear images provide improved personalization and recommendations for footwear. This study proposes a method to design customized shoes for foot conditions by leveraging AI-generated designs, biomechanical analyses, and material optimization.


12 November 2024 in IEEE Access
Verified
Journal Article
Open access
JDJaydeep Dey
Jaydeep Dey
PAP. Anandan
P. Anandan
SRSonaa Rajagopal
Sonaa Rajagopal
MMMuralikrishnan Mani
Muralikrishnan Mani
Polarized images, which record the polarization characteristics of light, are becoming increasingly important in a variety of applications like remote sensing, medical imaging, and target detection. Their ability to offer additional information beyond traditional intensity-based images makes them valuable in situations where conventional imaging methods are lacking. However, the use of polarized images for tasks such as target detection presents challenges due to the limited availability of datasets, resulting in subpar performance in deep learning algorithms. Traditional methods for improving the quality of polarized images often involve noise reduction techniques, but these approaches may not fully exploit on the potential of deep learning algorithms due to limited dataset access. To get over this restriction and improve the performance of deep learning models on polarised images, new approaches are required. In this study, a new approach is proposed to address the challenges linked with polarized images by harnessing the capabilities of GAN and deep learning models. Specifically, the MIRNet CNN algorithm is utilized to denoise enhanced polarized datasets produced by GANs. By training the deep learning model on these enhanced datasets, the aim is to boost the performance of subsequent tasks like target detection. The study demonstrates the efficacy and efficiency of this novel approach for bettering the polarised image performance of deep learning models, particularly the MIRNet and YOLOv8 models. Through the use of GAN-generated enhanced datasets, there is a notable enhancement in the accuracy of target detection utilizing YOLOv8. This highlights the potential of this approach not only in target detection but also in various other fields that rely on precise object detection and image denoising utilizing polarized images.