TY - GEN
T1 - Detection of Anomalies in Human Action Using Optical Flow and Gradient Tensor
AU - Mishra, Soumya Ranjan
AU - Mishra, Tusar Kanti
AU - Sarkar, Anirban
AU - Sanyal, Goutam
N1 - Publisher Copyright:
© 2020, Springer Nature Singapore Pte Ltd.
PY - 2020
Y1 - 2020
N2 - In this work, we present a tensor-based motion descriptor by combining both optical flow (OF) and histogram of oriented gradient (HOG) information from the video data to detect anomalous events. New combined aggregation method is proposed based on tensor descriptors. In video, motion is represented by polynomial coefficient and these coefficients approximate the optical flow (OF) and histogram of gradient (HOG) of video also used to represent the accumulated data. The coefficients are generated by projecting the motion vector on Legendre polynomials, and then sequence of coefficients are combined by using orientation tensors. In this paper, we have combined both tensor descriptors OF and HOG to capture the moving patterns in the video. We have trained the sequence of video containing only normal events by using SVM, and in testing phase, moving pattern of each region of the frame is compared with trained video to detect any types of anomaly events in the video. The proposed motion descriptor is evaluated on UCSD anomaly action dataset using SVM classifier and shows interesting results with very good accuracy.
AB - In this work, we present a tensor-based motion descriptor by combining both optical flow (OF) and histogram of oriented gradient (HOG) information from the video data to detect anomalous events. New combined aggregation method is proposed based on tensor descriptors. In video, motion is represented by polynomial coefficient and these coefficients approximate the optical flow (OF) and histogram of gradient (HOG) of video also used to represent the accumulated data. The coefficients are generated by projecting the motion vector on Legendre polynomials, and then sequence of coefficients are combined by using orientation tensors. In this paper, we have combined both tensor descriptors OF and HOG to capture the moving patterns in the video. We have trained the sequence of video containing only normal events by using SVM, and in testing phase, moving pattern of each region of the frame is compared with trained video to detect any types of anomaly events in the video. The proposed motion descriptor is evaluated on UCSD anomaly action dataset using SVM classifier and shows interesting results with very good accuracy.
UR - https://www.scopus.com/pages/publications/85075721796
UR - https://www.scopus.com/pages/publications/85075721796#tab=citedBy
U2 - 10.1007/978-981-13-9282-5_53
DO - 10.1007/978-981-13-9282-5_53
M3 - Conference contribution
AN - SCOPUS:85075721796
SN - 9789811392818
T3 - Smart Innovation, Systems and Technologies
SP - 561
EP - 570
BT - Smart Intelligent Computing and Applications - Proceedings of the 3rd International Conference on Smart Computing and Informatics, SCI 2018
A2 - Satapathy, Suresh Chandra
A2 - Bhateja, Vikrant
A2 - Mohanty, J.R.
A2 - Udgata, Siba K.
PB - Springer
T2 - 3rd International Conference on Smart Computing and Informatics, SCI 2018
Y2 - 21 December 2019 through 22 December 2019
ER -