Flight Safety - Alcohol Detection assisted by AI Facial Recognition Technology

Abstract

The Federal Aviation Administration’s (FAA) “Bottle to Throttle” rule requires that a pilot may not use alcohol within 8 hours of a flight and cannot have a blood alcohol content above 0.04% to ensure safety. However, a pilot could still feel intoxicated, even after the 8-hour window has expired. Some studies have shown decrements in pilot performance after 10 hours of drinking alcohol using a flight simulator. According to the FAA, aviation employees performing safety-sensitive functions are subject to random alcohol testing. Usually, a screening test and a confirmation test are administered to determine a prohibited alcohol concentration of a pilot. When a screening test is above the allowed limit, a confirmation test is conducted after 15 minutes. If the test can be conducted in a rapid and accurate manner before all pilots’ flights, the probability that a pilot is still under influence of alcohol, or drugs will be dramatically reduced. This project aims to develop an AI facial recognition technique using Human-Centered Computing (HCC) to more accurately identify if the pilot is still under influence of alcohol, instead of using the uniformed standard of 0.04% blood alcohol that may not apply for each individual.Rationale1) Blood alcohol level has a strong negative relationship with cognitive/ flight performance. For example, if an individual pilot is under the influence of alcohol use, their cognitive performance will be significantly impacted, as well as their flight performance will be significantly impacted.2) If an individual pilot is under the influence of alcohol use, the associated minor characteristics expressed on their faces. Those characteristics may not be able to be identified by human beings but can be detected by the AI facial recognition technique using HCC.3) Considering heterogeneity, the uniformed standard of 0.04% blood alcohol after 8 hours from drinking may not accurately reflect each individual pilot’s readiness to fly. The trained AI can detect whether an individual pilot is still under the influence of alcohol, regardless of blood alcohol concentration, which can serve as a rapid census test to ensure pilots’ safety to fly.MethodologyTask 1: Train the AI classification model by publicly available datasets of alcohol users’ facial images. Data source: Collect publicly available datasets containing facial images of individuals in various states of alcohol consumption (sober, mildly intoxicated, and heavily intoxicated). Images will be preprocessed using facial landmark detection and attention-based image segmentation to isolate key regions such as the eyes, nose, mouth, and skin tone changes that are sensitive to alcohol-induced variations. Model training: A deep convolutional neural network will be trained to classify the images and learn the relationship between facial features and alcohol-induced influences.Task 2: Examine the AI Facial Recognition algorithm among healthy average people (n=20) before and after certain amount of alcoholic beverage consumption. Specifically, the cognitive performance will be examined through a series of cognitive tasks focusing on deductive reasoning, working memory, concentration, associate-learning and spatial planning. The data will be used to fine-tune the AI facial recognition algorithm to more accurately identify alcohol influence. Task 3: Examine the AI facial recognition algorithm among pilots using flight simulator. The parameters of performance will be extracted from the simulator both before and after pilots’ alcohol consumption (n=20). The data will be used to customize the AI algorithm to customize it to pilots’ readiness to fly.Experiment ProceduresFor an average person, the capability of absorbing alcohol is 0.015 g/100ml/hour. In most healthy people, blood circulates through the body in 90 seconds, thereby allowing alcohol to affect cognitive function immediately, but full effects of a drink are felt within 15 to 45 minutes depending on the speed of absorption. In both experimental, each participant will spend 0.5-hour drinking, so the blood alcohol concentration will be around 0.018 % when they start their performances. Since participants’ alcohol absorbing rate varies (e.g., some participants may consume refreshments when drinking), their blood alcohol concentration will also be detected at several time points during the experiment by an alcohol breathalyzer tester as a reference.Significance1) Using HCC AI facial recognition, the alcohol test can be conducted on all pilots by simply scanning their faces, which improves flight safety compared to traditional spot-checking approaches.2) the AI facial recognition can also serve as the census screening test, before confirmation test by breath-testing device.3) the technology can more accurately identify if the pilot is still under influence of alcohol use, instead using the uniformed standard of 0.04% blood alcohol may not apply for each individual.

Keywords: flight safety, public safety, AI, safety policy, human centered computing

DOI: 10.54941/ahfe1006496