|This article applies to:||
What we will cover
You will learn how associations enable you to model entity relationships in your Alegion ontology, and make it simple for annotators to capture these relationships in video and images.
Associations are one of the lesser known, but more powerful features of Alegion video and image annotation. Let’s take a look at how associations capture interactions among labeled entities.
For our example, we’ll use a video use case in which entities are labeled and classified in a retail checkout scenario, and associations are used to define how those entities relate to each other. This fictitious use case is similar to several real world projects at Alegion. Associations are also commonly used in cases like skeletal keypoint annotation where multiple entities need to be grouped into a larger whole.
In this video, we have localized the body, head, and hands of both the cashier and the shopper. However, we also need to know what entities are associated with each person. In this case this includes the body parts, but also entities like shopping items that might be associated with different persons over time.
In other systems, relationships between entities might be inferred by containment by a localization, such as a bounding box, or set using entity ID fields or other methods. However, these techniques are not always reliable and can be time consuming and error prone. At Alegion we consider these methods inferior workarounds. So, we built in associations as a core part of the video and image ontology model we employ.
So, associating the heads and hands with the person is very simple and efficient. For each entity type, the correct associations are clearly defined and easily accessible.
For relationships that change over time, the model easily adapts. Here, for example, I’ll change the association to reflect that the shopping item has been passed from the cashier to the shopper.
These associations are defined in your ontology. Once you have entities defined, click to add associations between them. Once task are complete, relationships are clearly defined in the json output.
Associations capture the real world relationships between your labeled entities while enhancing efficiency and quality.