Object detection to locate and label signs and hazards in each frame is correct because it detects and labels multiple items and returns bounding boxes that show where each sign or hazard appears within a video frame.

Object detection models output class labels and bounding boxes for every detected object in each frame which supports real time driver assistance and can be combined with tracking to follow objects across frames. Bounding boxes are often sufficient for navigation decisions and they usually require less computation and lower latency than pixel level masks.

Amazon Comprehend is a text analytics service and it cannot analyze visual content so it cannot detect road signs or hazards in camera frames.

Image segmentation with pixel-level masks for every object in each frame gives precise object boundaries but it increases processing complexity and latency which is often unnecessary when bounding boxes meet the system requirements.

Simple image classification of each frame produces a single overall label for the image and it cannot locate or distinguish multiple objects within a frame so it does not satisfy the requirement to indicate where items appear.

Full AWS Practitioner Certification Question