La localisation d'objets fait référence à la vision par ordinateur task of identifying and precisely locating objects within digital images or video streams. This process typically involves classifying the objects present in the visual input and drawing bounding boxes around them to indicate their locations. Object localization is a critical component of many applications in intelligence artificielle and apprentissage automatique, particularly in fields such as véhicules autonomes, robotics, and surveillance systems.
The techniques used for object localization often rely on deep learning models, particularly Réseaux de neurones convolutifs (CNNs). These models are trained on large datasets that contain labeled images, allowing them to learn the features and characteristics of various objects. During inference, the model processes new images and produces predictions that include both the class of the object and its coordinates within the image.
Common algorithms used for object localization include the Region-based CNN (R-CNN), which segments regions of interest from an image and classifies them separately, and the YOLO (You Only Look Once) model, which performs localization and classification in a single pass, making it particularly suitable for real-time applications.
En plus de ses applications dans systèmes autonomes, object localization is also used in augmented reality, image retrieval, and content-based image analysis, showcasing its versatility across different technological domains.