Szenenverständnis refers to the ability of künstliche Intelligenz (AI) systems to interpret and analyze visual information from the world around them. This involves not just identifying objects within an image or video, but also understanding their spatial relationships, actions, and context within a scene.
At its core, scene understanding combines various techniques from computer vision, der Verarbeitung natürlicher Sprache, and machine learning. For example, when a self-driving car navigates through a city, it must recognize pedestrians, other vehicles, traffic signs, and obstacles while also understanding their movements and interactions. This requires a sophisticated level of perception that goes beyond simple recognition.
Zu den häufigen Aufgaben im Zusammenhang mit Szenenverständnis gehören:
- Objekterkennung: Identifikation und Lokalisierung von Objekten innerhalb eines Bildes.
- Semantische Segmentierung: Assigning a label to every pixel in an image, effectively categorizing different regions based on the objects present.
- Instanzsegmentierung: Unterscheidung zwischen einzelnen Instanzen desselben Objekts innerhalb einer Szene.
- Aktionserkennung: Verständnis dafür, welche Aktionen stattfinden und wer sie ausführt.
- Szene Klassifikation: Categorizing an entire image into a specific label or class, such as ‘beach’, ‘forest’, or ‘urban area’.
Szenenverständnis hat zahlreiche Anwendungen, einschließlich autonome Fahrzeuge, robotics, augmented reality, and surveillance systems. As AI technologies continue to evolve, improving scene understanding capabilities will enhance how machines interact with and respond to their environments.