Data has been collected from 200 taxis across eight locations nationwide, and as of October 2024, the total number of recorded instances has reached 226,000, including 62,000 records from single-camera units and 164,000 records from dual-camera units (some of which are unpublished).

  Moving forward, we plan to continue accumulating data from dual-camera units, which allow for monitoring drivers' eye movements and facial orientations. This will ensure that data remains relevant and reflective of the latest traffic conditions.

  The Near-Miss Database categorizes and records close-call incidents by risk level before they escalate into actual accidents. For every single accident, there are typically 10 high-level near-miss incidents, 50 medium-level, and 140 low-level incidents recorded, along with 900 instances of regular driving data.

  To achieve zero accidents, it is essential to analyze the wide range of near-miss incidents across different factors and situations, using them as a basis to develop preventative measures.

  The Near-Miss Database classifies near-miss incidents into approximately 280 categories, including target, accident type, location, legal violations, road conditions, and more. For example, statistical analysis can be conducted to determine the types of near-miss incidents associated with specific targets, providing detailed insights into accident patterns.

  Police data includes only personal injury accidents, whereas the Near-Miss Database encompasses a broader range, including property damage accidents, minor accidents not classified as property damage, and a large volume of near-miss incidents.

  While the overall distribution of accident types is similar across both datasets, there are some differences. For instance, there is a slightly higher proportion of incidents classified as collisions during overtaking. This may be due to categorizing pre-incident behaviors, such as cut-ins that can lead to rear-end collisions, as overtaking-related incidents.