MotoSafety: Edge-AI Achieves 94.97% Accuracy in Two-Wheeler Collision Risk Assessment Under Time Pressure
MotoSafety, a novel edge-AI framework, forecasts the likelihood of collisions for two-wheeled vehicles with an impressive accuracy of 94.97%. This innovative system utilizes a comprehensive dataset that includes over 129,000 labeled multivariate time-series sequences derived from 153 simulator rides, involving 51 participants under varying time-pressure scenarios—no, low, and high. It analyzes 64 features related to vehicle dynamics, control inputs, proximity, and behavioral infractions. MotoSafety surpasses ten benchmark models, such as TimesNet and LLM4TS, achieving a ROC AUC of 99.33% and a forecasting error of 0.039 MSE and 0.094 MAE, which is 4.4 times lower than Time-LLM and iTransformer. With a compact architecture of just 1.15M parameters and 0.135 ms latency, it is ideal for low-cost CPU hardware deployment, tackling essential safety issues for powered two-wheeler riders in low- and middle-income nations.
Key facts
- MotoSafety is a novel edge-AI architecture for two-wheeler collision risk assessment.
- The dataset includes over 129,000 labeled multivariate time-series sequences.
- Data was collected from 153 simulator rides by 51 participants.
- Subjects were tested under No, Low, and High Time Pressure conditions.
- 64 features were captured, covering vehicle dynamics, control inputs, proximity, and behavioral violations.
- MotoSafety achieves 94.97% accuracy and 99.33% ROC AUC.
- It outperforms ten baselines, including TimesNet and LLM4TS.
- With 1.15M parameters and 0.135 ms latency, the model is deployable on low-cost CPU hardware.
Entities
—