This page contains relevant information for future outdoor object-based localization methods by using landmarks:

• A mindmap as overview of proposed attributes for object description, georeferencing for the creation of future datasets.

• An object catalog as table with object-classes, attributes and advice for object-based localization suitability.

• • Download complete object catalog in excel format

Download the complete object catalog in excel format

Assumptions for calculating the object localization suitability

Important: All referenced sources of object attributes can be found in the Microsoft Excel table.

• A frontfacing camera is mounted in a car and driving along a straight road.
• While driving towards the object, a distance towards the object d = [-300 to 0] was assumed.
• Distance of the camera to the right side of the road = 1.75 m
• Germany has a total street length in = 627000 km
• Camera Z height (above the street) = 1.2 m
• Area of Germany = 357386 km²

object localization suitability calculatoin

The object localization suitability calculated by normed object projection on image plane, probability of occurence and persistence.

1. The real world object is projected via pinhole model to a camera image plane. The object area normalization on image plane is done via the integration of the object's projected area along the longitudinal distance $d_{x} \in [d_{min},d_{max}]$ to a frustum and its division through the distance:

$$ a_{img,norm} = \frac{1}{d_{max}-d_{min}} \cdot \int_{d_{min}}^{d_{max}} a_{img}(x_{veh}) * dx $$

2. The attribute probability of occurrence (PO) in objects per kilometre beside the roads.
3. The persistence of objects correlate with the update cycle of map.

Finally the object localization suitability is calculated:

$$ suitability = a_{img,norm} \cdot PO \cdot PS $$

Table with Object Attributes and Localization Suitability