What is an object detection?
One of the key aspects of the Mapillary platform is the computer vision technology embedded into our imagery processing pipeline. Using a method called semantic segmentation, we have trained an algorithm to assign a categorical label to each pixel making up a Mapillary image.
From there, we can discern that adjacent pixels assigned to the same category correspond to a given object, hence the term object detection. In short, an object detection is a grouping of pixels within an image that are believed to be a part of the same entity.
What types of object detections does Mapillary generate?
The above image provides a good visual of Mapillary's object detection capabilities. You can see that a majority of the image has overlaid object detections!
Chances are, if it's an object or characteristic typically associated with street-level imagery, Mapillary has a detection for it. Driveways, birds, pot holes, motorcyclists, tunnels, mountains—these are all examples of Mapillary object detection classes.
💡Complete list of Mapillary's 150 object detection classes
Additionally, it's important to note that a subset of our object detections are used to generate map features, which we group into point features and traffic sign classes. If you're not familiar with map features and their relation to object detections, make sure to check out our article on map features to learn more.
💡Complete list of Mapillary's 40 point features
💡Complete list of Mapillary's 1,500+ traffic sign classes
How can I view object detections?
Object detections can be seen by clicking into an image on the Mapillary web app, however first you must toggle them on. To learn how to view object detections in the web app, see our article on exploring object detections.
Only object detections associated with point and traffic sign map features can be viewed via the web app at this time.
How can I integrate Mapillary object detections into my own workflows?
To query, download, and process Mapillary object detections, you must use the Mapillary API.
Currently, the API supports two different end points for object detections—you can retrieve object detections via an image id or a map feature id.
Full documentation of the API is available here. To see an example on how to use the Mapillary API to obtain all object detections within a sequence, see the Mapillary API FAQ.
Comments
2 comments
What is the principle of triangulating the location of detected objects in several images and retrieving the real world location of objects?
Hi! Wikipedia has a great article on triangulation: https://en.wikipedia.org/wiki/Triangulation_(computer_vision)
This is the principle we use as well when triangulating object detections, in order to generate map features that represent the objects' real-world locations.
Please sign in to leave a comment.