Timestamp aware Aberrant Detection and Analysis
in Big Visual Data using Deep Learning Architecture
Science and Engineering Research Board (SERB): SERB/EEQ/2017/000673
Funding Agency: Science and Engineering Research Board, Department of Science and Technology (SERB-DST, 2018)
Principal Investigator: Dr. Santosh Kumar Vipparthi
JRF/Ph.D. Scholar: Kuldeep Marotirao Biradar
Introduction: The proposed system removes the onus of detecting aberrance situations from the manual operator;
and rather, places it on the video surveillance system.
The present technologies are fails to recognize aberration in video sequences. These aberrances
occur over a small-time window. Thus, recognizing with its timeframe from a big visual data is really
challenging task. Hence, our focus is on problems, where we are given a set of nominal training
videos samples. Based on these samples need to determine whether or not a test video contains an
aberration and what instant it occurs. Similarly, we aim to significantly reduce the time and human
effort by automating the task and improving the accuracy by recognizing aberrances with its
timestamp. Further, exploit the aberrance activity of the object by modeling the rich motion
patterns in selected region, effectively capturing the underlying intrinsic structure they form in the
video. Implementation of this system can be beneficial for intelligent agencies, banks, departmental stores, traffic monitoring on highway,
airport terminal check-in, sports, medical field, and robotics etc.
PROJECT ACTIVITIES AND FINDINGS
Anomaly Detection in Traffic Videos
New Anamoly Dataset
A custom dataset was generated in a staged/controlled environment. We shot from four strategically placed cameras
simultaneously to capture multiple views of same scene. The videos were recorded at four different locations in different times
of the day. The scenes involve normal data, fight happening in different scenarios, snatching, kidnapping etc.
Scenes were shot indoor/outdoor, in natural light-artificial light, low light as well to cover illumination changes. The videos were recorded
from varied distances to capture subjects with varying size. Post processing yielded usable clips of approx. 90 minutes (90x60x30x4= 648000 frames).
Snippet for the same are depicted in the figure.
Murari Mandal, Lav Kush Kumar, Mahipal Singh Saran, Santosh Kumar Vipparthi, “MotionRec: A Unified Deep Framework for Moving Object
Recognition,” IEEE Winter Conference on Applications of Computer Vision (WACV), Snowmass Village, Colorado, US, 2020
Monu Verma, Santosh Kumar Vipparthi, Girdhari Singh, Subrahmanyam Murala, “LEARNet:
Dynamic Imaging Network for Micro Expression Recognition,” IEEE Transactions on image
processing, 2019 (Impact Factor 6.79).
Kuldeep Marotirao Biradar, Ayushi Gupta, Murari Mandal, Santosh Kumar Vipparthi,
"Challenges in Time-Stamp Aware Anomaly Detection in Traffic Videos,"
" In CVPR Workshops (CVPRW), Long Beach, California, US, 2019.
12.Monu Verma, Jaspreet Kaur Bhui, Santosh Kumar Vipparthi,
Girdhari Singh, "EXPERTNet: Exigent Features Preservative Network for Facial
Expression Recogntion," ACM 11th International Conference on Computer Vision, Graphics and
Image Processing (ICVGIP), Hyderabad, India, 2018.
15.Monu Verma, Prafulla Saxena, Santosh Kumar Vipparthi, Girdhari Singh, S K Nagar,
“DeFINet: Portable CNN Network for Facial Expression Recognition,”
IEEE International Conference on Information and Communication Technology for Competitive Strategies, 2019
Murari Mandal, Manal Shah, Prashant Meena, Sanhita Devi, Santosh Kumar Vipparthi, “AVDNet:
A Small-Sized Vehicle Detection Network for Aerial Visual Data,” IEEE Geoscience and Remote Sensing Letters, doi: 10.1109/LGRS.2019.2923564 (Impact Factor 3.534)