Page 23 - AI Ready – Analysis Towards a Standardized Readiness Framework

AI Ready – Analysis Towards a Standardized Readiness Framework

4 Use Case Analysis

For each use case, there are several sub-scenarios investigated.

For example, driver distraction detection [5] [18][26] under the topic of traffic safety is an
umbrella use case, under which there are several scenarios such as – drowsiness detection
[25], texting while driving, and detecting lane infractions. Similar to this transportation parent
use case, in the “accessibility” use case [2] [77], there are also different scenarios, each of which
has a specific focus for hearing-impaired groups and visually impaired users.

For each of these sub-scenarios, the actors could be specific to the scenario, for example, the
major type of sensor may be different in each scenario.

The classification of the source of data (SRC) [ITU-T Y.3172] into external and internal (to the
vehicle or system) can be used to derive additional metadata about the source. Roadside units
(RSU) can tell whether the lane discipline is correctly followed, and v2v sensors can tell if the
distance is maintained. The SINK [ITU-T Y.3172] could be integral (built-in) controls e.g. braking
systems [45] and accelerating systems in a vehicle, however, there could be vendor-provided
differentiated signal processing which is applied on such controls that are used to find if the
driver is distracted. Apart from these integral controls, another type of controls could be add-
ons provided by 3 parties. For example, Add-ons such as driver observation cameras [6] or
rd
rd
sensors may be purchased from 3 parties and installed in the vehicles by owners. In such cases,
Interoperability and integration aspects are critical for the overall use case.
Simulation of use case-specific scenarios is important to validate and certify the compliance of
various actors to local and global standards and regulations [30]. To make sure the end-user
has a common understanding, simulation setups, parameters, and boundary values may be
studied in Standards Developing Organizations (SDO).

Thus, standards and regulations studies related to the quantification of the specific actors
and behaviours including various simulation and validation scenarios along with compliance
thresholds are needed.

The application of AI in various scenarios brings the question of trust, explainability, traceability,
and accountability for decisions. Models and algorithms used in the use cases are required
to be trusted, explainable, and accountable. Deployment and hosting conditions may meet
specific requirements such as low latency in the vehicle, low memory footprint, and low power
consumption.

Mitigation of privacy concerns such as data handling of systems where there is personal data
involved, needs further study. E.g. The distraction detection system may use a camera, but the
system is designed to purge the visual data permanently within a regulated time period. The
non-visual data may still be stored (e.g. alarms and timestamps). However, there are regulations
such as sharing raw data with law enforcement agencies.

4.1 Use Case Summaries

The use case summaries below give specific details of the use case that are relevant for the AI
readiness study. The details of the use cases with notes on characteristics are in Appendix A:
“Detailed analysis of the use cases and AI impacts on the use cases”. In addition, Appendix B:
“Specific impacts of these characteristics on Standards Frameworks for AI readiness require

16