Kanimozhi Soundararajan
@skanimozhi@vit.ac.in
Assistant Professor (Sr. Gr) and School of Computer Science and Engineering
VIT University, Chennai Campus
EDUCATION
Ph.D (ICT)
M.Tech(IT)
B.Tech(IT)
RESEARCH, TEACHING, or OTHER INTERESTS
Multidisciplinary, Computer Engineering, Agricultural and Biological Sciences, Computers in Earth Sciences
Scopus Publications
Scopus Publications
S. Kanimozhi, A. Sasithradevi, and L. Sairamesh
Institute of Electrical and Electronics Engineers (IEEE)
Sports employ computer vision for a variety of applications, including interactive viewing, object interpretation, sports action analysis, and intelligent rule-based systems for better referee choices. The computer vision problem known as “sports action recognition” looks for interactions between players and related sporting items. Most of the existing solutions in sports action recognition are complex and incomplete as taking both playfield and non-play field objects with various actions for the same human posture into account. Hence, the proposed model focuses only on objects within the playfield and the right functional action that is performed by the human. We identify the play field using a quadrant-based density (QD) approach which locates the crowded region and then morphological operations are applied to spot only play region. The right action performed is identified using the logic-based query to the Sports Action Generation model (SAG) which is generated using the newly constructed sports ontology. The proposed semantic verification model using ontology typically offers more transparency compared to deep learning networks. It relies on explicit rules and logic that can be easily interpreted, enhancing interoperability and accuracy across systems. We use the sports video wild (SVW) and UCF-101 datasets to assess the performance of our proposed model. The proposed sports action recognition system achieves action identification in videos with a good accuracy of 96%.
S. Kanimozhi and L. Sairamesh
Institution of Engineering and Technology
Kanimozhi Soundararajan and Mala T
SAGE Publications
Computer vision in sport is a very interesting application. People spend a lot of time watching sports videos because this is one of the best field of entertainment. Sports video broadcasts generally take a lot of time, ranging from two to four hours. However, the interesting part happens for just a few minutes. Detecting the highlighted event in a sport will be useful for people who like to watch only the prominent events section instead of watching the whole video broadcast. Event detection will give precise details about the action that occurred for a particular time, but the detection of highlighted events is more complex. This is due to the fact that a sports video contains collections of events. Among them, segregation of the required event is a time-consuming process but it requires more knowledge about the sport as well as processing time. Hence, a novel work is proposed focused on identifying the location of the functional object using agglomerative clustering and annotating the event highlights automatically by means of the rule inference mechanism. The SHRED (Sports Highlight Recognition and Event Detection) system achieves an overall accuracy of about 97.38% relative to other state-of-art methods in event class annotation.
Kanimozhi Soundararajan, Anbarasi Soundararajan, and Sai Ramesh L
IOS Press
Anomaly detection is a challenging task in the surveillance system due to the factors like extracting appropriate features, inappropriate differentiation among the normal vs abnormal behaviours, the sparse occurrence of abnormal activities and environmental variations. In the dark environment, detection of human actions is still difficult as more features for recognizing the key point are not visible. Hence the proposed work is focused on overcoming the environmental variations task that too in a less bright environment by using thermal videos. Variations in the actions can be easily identified as it works on the property of infrared radiations. For recognizing actions, the skeleton-based approach is used as it helps with the joint-wise segregation of human parts, resulting in more accuracy. The motion pattern of humans in the thermal video is tracked to classify the level of abnormality.
S. Kanimozhi, T. Mala, A. Kaviya, M. Pavithra, and P. Vishali
Springer Singapore
Thaseena Sulthana, Kanimozhi Soundararajan, T. Mala, K. Narmatha, and G. Meena
Springer International Publishing
Kanimozhi S, Anbarasi S, and Mythili M
IOS Press
Recognizing human action in sports is difficult task as various sequences of activities involved in every scene. Identifying each action individually without overlapping of movements is a tedious process due to continuous change of frames within short duration. So proper tracking of human movements for each action is important. Hence new structure-based human action recognition and tracker technique (HART) is proposed. It uses joint trajectory images and visual feature to design each human action. At first, a structural based method employed to extract human skeleton data points from RGB (Red Green Blue) videos. Next, a Multitude Object Tracker (MOT) is proposed which uses the trajectory of human skeleton joints in an image space for identification of actions. Then, Histogram of Oriented Gradients (HOG) combined with Support Vector Machine (SVM) is applied to extract physical body shape and action information. Finally, the action label and interconnected keypoints in humans is jointly detected as end result. The proposed HART technique effectively performed well with the accuracy of about 82% over the other activity recognition methods.
S Kanimozhi, G Gayathri, and T Mala
IEEE
Real time object detection is one of the challenging task as it need faster computation power in identifying the object at that time. However the data generated by any real time system are unlabelled data which often need large set of labeled data for effective training purpose. This paper proposed a faster detection method for real time object detection based on convolution neural network model called as Single Shot Multi-Box Detection(SSD).This work eliminates the feature resampling stage and combined all calculated results as a single component. Still there is a need of a light weight network model for the places which lacks in computational power like mobile devices( eg: laptop, mobile phones, etc). Thus a light weight network model which use depth-wise separable convolution called MobileNet is used in this proposed work. Experimental result reveal that use of MobileNet along with SSD model increase the accuracy level in identifying the real time household objects.
RECENT SCHOLAR PUBLICATIONS
Publications
Kanimozhi, S., Gayathri, G., & Mala, T. (2019, February). Multiple Real-time object identification using Single shot Multi-Box detection. In 2019 International Conference on Computational Intelligence in Data Science (ICCIDS) (pp. 1-5). IEEE.
Kanimozhi, S., Mala, T., Kaviya, A., Pavithra, M., & Vishali, P. (2022). Key Object Classification for Action Recognition in Tennis Using Cognitive Mask RCNN. In Proceedings of International Conference on Data Science and Applications (pp. 121-128). Springer, Singapore.
Soundararajan, K. (2022). Sports highlight recognition and event detection using rule inference system. Concurrent Engineering, 1063293X221088353.
Sulthana, T., Soundararajan, K., Mala, T., Narmatha, K., & Meena, G. (2021, March). Captioning of Image Conceptually Using BI-LSTM Technique. In International Conference on Computational Intelligence in Data Science (pp. 71-77). Springer, Cham.