Abstract:Surveillance video analysis using automated AI-based techniques is a prominent research field with real-world applications. Several techniques aiming at recognizing activities, behaviour, and violent actions are present in literature. Real-world data analysis for violence detection is still a major challenge due to the limited datasets available for training with complex scenarios and varied scaling of objects performing different activities. In this paper, we present a novel deep learning model considering specially designed frame encoders for spatial feature extraction that are generalized towards many challenges, such as light conditions and indoor and outdoor scenarios. Furthermore, the spatial features in the stacked form are analyzed using a temporal deep learning model to observe and learn the temporal patterns dependencies by considering past and future information while predicting the violent or normal class. In the literature, violence is considered a binary classification of either fight or no fight. Different from these techniques, we present a multi-class classification of violent activities considering different types of human violence, such as assault, shooting, etc. The data for multi-class violence classification is extracted from the famous real-world anomaly detection UCF-crime dataset, where only six human-involved types of anomalies are considered for preparing training and testing sets. We report 41.1%, 48.85% accuracy for 16 and 8 frames in a single sequence. This data shows that more research is demanded to increase the robustness of deep models in violence detection. Finally, the proposed results against recent methods over violence detection datasets are marginally better, indicating the effectiveness of the proposed feature extraction and temporal learning mechanism.

Physical Violence Detection with Movement Sensors.

Mobile Neural Architecture Search Network and Convolutional Long Short-Term Memory-Based Deep Features Toward Detecting Violence from Video

Towards Real-world Violence Recognition via Efficient Deep Features and Sequential Patterns Analysis

Violence Behavior Detection Based on 3D-CNN

A Violence Detection Approach Based On Spatio-Temporal Hypergraph Transition

Emotion Recognition of Physical Activities for Health Monitoring Using Machine Learning

Combining OpenPose with BiLSTM for Violence Detection in Long-Term Care

Activity Recognition on an Accelerometer Embedded Mobile Phone with Varying Positions and Orientations

Recognition System of Human Daily Physical Activity Based on a 3D Acceleration Sensor

Violent Interaction Detection in Video Based on Deep Learning

Recognizing Violent Activity Without Decoding Video Streams

A Skeleton-based Approach for Campus Violence Detection

Efficient Human Violence Recognition for Surveillance in Real Time

Feature extraction for robust physical activity recognition

SOS: Real-time and Accurate Physical Assault Detection Using Smartphone.

A Wearable Inertial Sensor Approach for Locomotion and Localization Recognition on Physical Activity

Utilizing Deep Learning Models to Develop a Human Behavior Recognition System for Vision-Based School Violence Detection

An Overview of Violence Detection Techniques: Current Challenges and Future Directions

Real Time Violence Detection Based on Deep Spatio-Temporal Features.

Violence detection and face recognition based on deep learning

Body Motion Recognition Based on Acceleration Sensor