Highly efficient anomaly detection in traffic surveillance videos with optimized interference-tolerant fast convergence zeroing neural network

Giriprasad, S.

doi:10.1016/j.image.2026.117553

Highly efficient anomaly detection in traffic surveillance videos with optimized interference-tolerant fast convergence zeroing neural network

2026-05-23

0

PRE

AI

G

Giriprasad, S. *

R

Ravikumar, G.

S

Shanmugam Gokul

G

Govardhan, S. D.

DOI：10.1016/j.image.2026.117553

Original

Original request for help

Save

Abstract

En 中文

The process of recognizing anomalous activity involves spotting patterns, occurrences that deviate from regular flow. In the surveillance paradigm, such incidents could range from abuse to fights, car crashes, snatching. To overcome this issue, a Highly Efficient Anomaly Detection in Traffic Surveillance Videos with Optimized Interference-Tolerant Fast Convergence Zeroing Neural Network (HAD-TSV-ITF-CZNN) is proposed in this paper. Here, the video frames are taken from UCF Crime dataset. Then the video frames are given into the preprocessing stage. During pre-processing, each frame is resized, normalized, spatial augmentation using Sub aperture Keystone Transform Matched Filtering (SAKTMF). Then videos are fed into the feature extraction stage, where, anomalous spatiotemporal features, like object position, velocity, optical flow and crowd density are extracted using High Order Synchro extracting Transform (HOST). The extracted features are fed to the ITFCZNN for classifying the anomaly detection as normal, road accident and explosion. The efficiency of the HAD-TSV-ITF-CZNN technique is evaluated under some performance metrics: accuracy, precision, recall, f1-measure, AUC, error rate. The HAD-TSV-ITF-CZNN achieves 23.12%, 21.23% and 21.32% higher accuracy, 24.28%, 22.54% and 24.32% higher precision, 21.14%, 21.45% and 20.24% higher recall compared with existing models: anomaly recognition from surveillance videos utilizing 3D convolution neural network (ARF-SV-CNN), CNN features by bi-directional LSTM for real-time anomaly detection in surveillance networks (CNN-LSTM-ADSN) and video anomaly detection system utilizing deep convolutional with recurrent methods (VAD-DC-RM) respectively.

Keywords:

Interference-tolerant fast convergence zeroing neural network

Musical chairs optimization algorithm

Sub aperture Keystone Transform Matched

Filtering

video content analysis

Highly efficient anomaly detection in traffic surveillance videos with optimized interference-tolerant fast convergence zeroing neural network

Abstract

Journal

Organization

Cited Papers

Citing Papers