The core idea behind surrogate measures of safety (SMoS) is that the accident risk could be measured through indicators other than accident counts.

The main limitations of accident counts as a safety indicator are:

• Accidents are random and low in number; years of data are necessary for any meaningful analysis;
• Accidents are often under-reported in the official records;
• Accident reports are filled in post-factum, potentially missing important details;
• It is unethical to passively wait for accidents to happen as a part of the routine in road safety work.

Since 1970, our group has been working on the methods and tools to measure safety through traffic conflicts. Traffic conflict is an event, in which the actual collision has been avoided, yet the reasonable safety margins have been compromised. On the severity scale between normal interactions and accidents, traffic conflicts fall into the transition area, being still very similar to accidents (thus the process and contributing factors can be studied in real time), but also being much more frequent compared to the real collisions.

During 1970-1990s, the Swedish Traffic Conflict Technique was developed and validated. The technique is based on human observations in traffic or from video. It relies on Time-to-Collision and Conflicting Speed to grade the severity of the situations and select serious conflicts. It has been repeatedly shown, based on large datasets, that serious conflicts are strongly correlated with police-reported accident counts.

Starting from 2000, the focus shifted towards automation of the conflict detection and analysis, leading to tight co-operation with the researchers on image and video analysis and machine learning.

More about our research on video analysis in traffic