Table 6.

Summary of literature on induvial user-based HAR.

Ref.	Year	Description
[77]	2018	SVM and the N-cut algorithm were used to label video segments, and the CRF was used to detect anomalous events.
[78]	2018	A deep convolutional framework was used to develop a unified framework for detecting abnormal behavior with LSTM in RGB images. YOLO was used determine the action of individuals in video frames and then VGG-16 classify them.
[79]	2018	Proposed a HOME FAST spatiotemporal feature extraction approach based on optical flow information to detect anomalies. Proposed approach obtained low-level features with KLT feature extractor and supplied to DCNN for categorization.
[80]	2019	Proposed an algorithm used adaptive transformation to conceal the affected area and the pyramid L-K optical flow method to extract abnormal behavior from videos.
[81]	2019	By combining extracted hidden patterns of text with available metadata, a deep learning architecture RNN was proposed to detect abusive behavioral norms.
[10]	2019	SVM was used to determine abnormal behavior using extracted feature vectors and vector trajectories from the computed optical flow field of determined joint points with LK method.
[82]	2019	The proposed LSTM-FCN detects aggressive driving sessions as time series classification to solve the problem of driver behavior.
[83]	2019	A method that combined CNN with HOF and HOG was proposed to detect anomalies in surveillance video frames.
[84]	2020	A deep learning model was used to detect abnormal behavior in videos automatically, and experiments with 2D CNN-LSTM, 3D CNN, and I3D models were conducted.
[85]	2020	Propose to do instance segmentation in video bytes and predicting the actions with the help of DBN based on RBM. Aimed to present an implementation of an algorithm that can depict anomalies in real time video feed.
[86]	2021	Proposed a method for detecting abnormal behavior that is both accurate and effective. VGG16 network transferred to full CNN to extract features. Then LSTM is used for prediction at that moment.
[87]	2021	Proposed a method in the ABAW competition that used a pre-trained JAA model and AU local features.
[88]	2021	Proposed a strategy for recognizing and detecting anomalies in human actions and extracting effective features using a CPRTSA based Deep Maxout Network.
[89]	2021	The algorithm was classified into two types. The first employs data mining and knowledge discovery, whereas the second employs deep CNN to detect collective abnormal behavior. Researcher planned variation of DBSCAN, kNN feature selection, and ensemble learning for behavior identification.
[90]	2021	Residual LSTM was introduced to learn static and temporal person-level residual features, and GLIL was proposed to model person-level and group-level activity for group activity recognition.