DETECTION OF POTENTIALLY RISKY DRIVING SCENES AND IDENTIFICATION OF ASSOCIATED RISK FACTORS
- Delivery
- Available on this site
- Format
- Price
- Non-members (tax incl.)：￥1,100 Members (tax incl.)：￥880
- Publication code
- 20219059
- Paper/Info type
- Other International Conferences
- Pages
- 1-5(Total 5 p)
- Date of publication
- Sep 2021
- Publisher
- JSAE
- Language
- English
Detailed Information
| Author(E) | 1) Masahiro Ichiki, 2) Chiyomi Miyajima, 3) Alexander Carballo, 4) Kazuya Takeda |
|---|---|
| Affiliation(E) | 1) Nagoya University, 2) Daido University, 3) Nagoya University, 4) Nagoya University |
| Abstract(E) | We propose a method to simultaneously detect potentially risky driving scenes and identify the associated risk factors, using a deep neural network (DNN) and human risk perception. We first categorize possible risk factors in driving scenes into environmental factors and ego-vehicle behavior factors, and the environmental factors are then further classified into dynamic and static factors. The ground-truth of risk levels of driving scenes and associated risk factors are obtained using subjective risk perception data provided by human annotators. Semantic segmentation and object detection methods are applied to the video images to obtain information about risk factors present in the surrounding environment, such as the presence of dynamic obstacles and static road information. These features are then fed into a DNN, which automatically detects potentially risky scenes while simultaneously identifying the associated risk factors. We also analyze differences in the risk perception of driving instructors and regular drivers, using their annotation data. Our experimental results show that risky scene detection and risk factor identification performance are both improved by the simultaneous method, compared with the performance of methods which use independent DNNs trained to perform these tasks separately. We also found that regular drivers tend to select static objects and ego-vehicle behavior as risk factors more often than driving instructors, who tend to select dynamic features as risk factors, and who perceive potentially risky situations earlier than regular drivers, even when no risk factors are currently present. |
