A Sensor Network Approach for Violence Detection in Smart Cities Using Deep Learning

Area of Science:

• Computer Science
• Artificial Intelligence
• Urban Planning

Background:

• Citizen safety is crucial for urban quality of life.
• Smart city surveillance generates massive data, posing processing challenges.
• Existing real-time violence detection methods require high computational power, unsuitable for embedded systems.

Purpose of the Study:

• To propose an efficient method for automatic violent behavior detection in video sensor networks.
• To enable real-time violence detection on resource-constrained embedded architectures like Raspberry Pi.
• To overcome the limitations of existing high-processing-power solutions.

Main Methods:

• A computationally effective cascaded approach separating temporal and spatial information processing.
• Utilizing a deep neural network followed by a time domain classifier.
• Feeding the deep neural network with motion vector features extracted directly from MPEG encoded video streams.

Main Results:

• The proposed algorithm achieves real-time processing on a Raspberry Pi-embedded architecture.
• The method demonstrates state-of-the-art performance in violent behavior detection.
• The approach is suitable for low computational resource environments.

Conclusions:

• The developed algorithm offers an efficient solution for real-time violent behavior detection in smart city video surveillance.
• This method is optimized for embedded systems, addressing the limitations of current high-resource solutions.
• The findings contribute to enhancing citizen safety through advanced, accessible video analytics.