Efficient abnormal event detection in video using deep attention based bidirectional lstm with a mayfly optimization

Abnormal event detection is a challenging issue in video surveillance, and it is quite necessary for detecting suspicious behaviour in the normal video data. Detecting abnormalities in video is very crucial and the application ranges from automatic control of quality to visual surveillance data. This paper presented efficient abnormal event detection in video utilizing deep attention based bidirectional LSTM (Long Short Term Memory) with a Mayfly optimization. Initially, the key frames of input video are selected utilizing threshold based discrete wavelet transform. In the second stage, Kernel Entropy Component Analysis (KECA) is used for decreasing the dimensionality. In the third stage, optimal weighted bilateral filtering is utilized for removing the unnecessary noises. In the next stage, a hybrid dual tree Gabor transform is utilized for the effective feature extraction. Afterwards, the Farne back optical methodology is incorporated to estimate the motion in the video sequence. In the final stage, deep attention based bidirectional LSTM with a mayfly optimized model effectively detects the abnormal events. This presented methodology effectively detects normal and abnormal events and it is implemented in PYTHON platform. The performance of the proposed approach is tested on QMUL and UCF datasets. The experimental outcomes of the presented methodology proved that the presented work is significantly better in terms of various effective performance measures like accuracy, AUC (Area Under Curve), execution time and ROC (Receiver Operating Characteristics) measures. The proposed approach achieved the improved outcomes in terms of accuracy as (94.19%) for QMUL dataset, and (93.60%) for UCF dataset.