At a marathon, each runner has a race ID printed on the front of their shirt. The company must extract those race IDs from runner images with the least operational overhead. Which option meets this requirement?

Amazon Rekognition is the most suitable choice because its DetectText API can directly extract text from images, including race IDs from runner shirts, with minimal operational overhead. It's a fully managed service, requiring no infrastructure setup or model training. Developing and hosting a custom CNN would involve significant development effort, data labeling, model training, and infrastructure management, leading to high operational overhead. SageMaker Object Detection is designed for identifying and localizing objects within images, not specifically for text extraction, and would still require training and management, increasing overhead compared to Rekognition. Amazon Lookout for Vision is specialized for detecting defects in industrial products, making it inappropriate for text extraction from runner images.