A CNN-RNN Combined Structure for Real-World Violence Detection in Surveillance Cameras
Bibliographic record
Abstract
Surveillance cameras have been increasingly used in many public and private spaces in recent years to increase the security of those areas. Although many companies still recruit someone to monitor the cameras, the person recruited is more likely to miss some abnormal events in the camera feeds due to human error. Therefore, monitoring surveillance cameras could be a waste of time and energy. On the other hand, many researchers worked on surveillance data and proposed several methods to detect abnormal events automatically. As a result, if any anomalous happens in front of the surveillance cameras, it can be detected immediately. Therefore, we introduced a model for detecting abnormal events in the surveillance camera feed. In this work, we designed a model by implementing a well-known convolutional neural network (ResNet50) for extracting essential features of each frame of our input stream followed by a particular schema of recurrent neural networks (ConvLSTM) for detecting abnormal events in our time-series dataset. Furthermore, in contrast with previous works, which mainly focused on hand-crafted datasets, our dataset took real-time surveillance camera feeds with different subjects and environments. In addition, we classify normal and abnormal events and show the method’s ability to find the right category for each anomaly. Therefore, we categorized our data into three main and essential categories: the first groups mainly need firefighting service, while the second and third categories are about thefts and violent behaviour. We implemented the proposed method on the UCF-Crime dataset and achieved 81.71% in AUC, higher than other models like C3D on the same dataset. Our future work focuses on adding an attention layer to the existing model to detect more abnormal events.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".