Related Experiment Video
Updated: Jul 30, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
591
Integrating Spatial and Temporal Information for Violent Activity Detection from Video Using Deep Spiking Neural
Xiang Wang1, Jie Yang1, Nikola K Kasabov2
1Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University, Shanghai 200400, China.
Sensors (Basel, Switzerland)
|May 13, 2023
Summary
This study introduces SpikeConvFlowNet, a deep learning model for detecting workplace violence in videos. It efficiently analyzes visual and motion data, offering a promising solution for real-time security monitoring systems.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Workplace violence, particularly in healthcare settings, poses significant safety concerns.
- Real-time visual monitoring of extensive video data is resource-intensive and impractical.
- Automatic and efficient detection of violent activities is crucial for enhanced security.
Purpose of the Study:
- To propose an efficient deep learning architecture for automatic video-based violent activity detection.
- To address the limitations of manual video surveillance in identifying violent incidents.
- To develop a solution suitable for resource-constrained monitoring systems.
Main Methods:
- A two-stream deep learning architecture, SpikeConvFlowNet, was developed.
- The model processes both RGB frames and optical flow data to capture spatiotemporal features.
- Spiking neural networks with convolutional and pooling layers were employed for feature extraction and temporal analysis.
Main Results:
- SpikeConvFlowNet demonstrated reduced parameters and improved inference efficiency compared to state-of-the-art methods.
- The approach achieved competitive accuracy with significant gains in processing speed.
- The model effectively extracts spatial and temporal features, including critical motion information.
Conclusions:
- SpikeConvFlowNet offers an efficient and effective solution for video violent activity detection.
- The architecture is well-suited for embedded devices requiring high processing speeds and low power consumption.
- This method enhances the potential for real-time security applications in various environments.
