A New Video-Based Crash Detection Method: Balancing Speed and Accuracy Using a Feature Fusion Deep Learning Framework
Bibliographic record
Abstract
Quick and accurate crash detection is important for saving lives and improved traffic incident management. In this paper, a feature fusion-based deep learning framework was developed for video-based urban traffic crash detection task, aiming at achieving a balance between detection speed and accuracy with limited computing resource. In this framework, a residual neural network (ResNet) combined with attention modules was proposed to extract crash-related appearance features from urban traffic videos (i.e., a crash appearance feature extractor), which were further fed to a spatiotemporal feature fusion model, Conv-LSTM (Convolutional Long Short-Term Memory), to simultaneously capture appearance (static) and motion (dynamic) crash features. The proposed model was trained by a set of video clips covering 330 crash and 342 noncrash events. In general, the proposed model achieved an accuracy of 87.78% on the testing dataset and an acceptable detection speed (FPS > 30 with GTX 1060). Thanks to the attention module, the proposed model can capture the localized appearance features (e.g., vehicle damage and pedestrian fallen-off) of crashes better than conventional convolutional neural networks. The Conv-LSTM module outperformed conventional LSTM in terms of capturing motion features of crashes, such as the roadway congestion and pedestrians gathering after crashes. Compared to traditional motion-based crash detection model, the proposed model achieved higher detection accuracy. Moreover, it could detect crashes much faster than other feature fusion-based models (e.g., C3D). The results show that the proposed model is a promising video-based urban traffic crash detection algorithm that could be used in practice in the future.
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) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".