Reproduce “Anomalous Motion Detection on Highway Using Deep Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Harpreet Singh, Emily M. Hand, Kostas Alexis
Researchers from the University of Nevada, Reno, introduced the Highway Traffic Anomaly (HTA) dataset, the first open-source dataset for visual anomaly detection in dynamic highway environments from a moving vehicle's perspective. They evaluated existing deep learning models on this dataset and found that a modified Predictive Coding Network (PredNet N+6) improved detection for several anomaly types, achieving an AUC of 0.828 for speeding motorcycles.
Key insights
- Existing deep learning anomaly detection models, effective in static camera settings, do not directly transfer well to dynamic environments with ego-motion due to continuously changing scenes and backgrounds.
- Discriminator outputs from models like CGAN, which are useful as anomaly heatmaps in static backgrounds, are less effective in dynamic highway scenes.
- Predicting further into the future (e.g., 6 frames ahead with PredNet) can amplify the effects of subtle anomalous motion, making it more discernible than short-term predictions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.