Traffic Risk Assessment: A Two-Stream Approach Using Dynamic-Attention
Bibliographic record
Abstract
The problem being addressed in this research is performing traffic risk assessment on visual scenes captured via outward-facing dashcam videos. To perform risk assessment, a two-stream dynamic-attention recurrent convolutional neural architecture is used to provide a categorical risk level for each frame in a given input video sequence. The two-stream approach consists of a spatial stream, which analyzes individual video frames and computes high-level appearance features and a temporal stream, which analyzes optical flow between adjacent frames and computes high-level motion features. Both spatial and temporal streams are then fed into their respective recurrent neural networks (RNNs) that explicitly models the sequence of features in time. A dynamic-attention mechanism which allows the network to learn to focus on relevant objects in the visual scene is added. These objects are detected by a state-of-the-art object detector and correspond to vehicles, pedestrians, traffic signs, etc. The dynamic-attention mechanism not only improves classification performance, but also provides a method to visualize what the network "sees" when predicting a risk level. This mechanism allows the network to implicitly learn to focus on hazardous objects in the visual scene. Additionally, this research introduces an offline and online model that differ slightly in their implementations. The offline model analyzes the complete video sequence and scores a classification accuracy of 84.89%. The online model deals with an infinite stream of data and produces results in near real-time (7 frames-per-second); however, it suffers from a slight decrease in classification accuracy (79.90%).
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 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| 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 | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".