Method and system for estimating the trajectory of an object on a map

An apparatus for estimating a trajectory of an object on a map, comprising:

• one or more processors;
• and
• memory including code that, when executed by the one or more processors, performs functions including:
• receiving a sequence of traces of the object on the map;
• each trace in the sequence of traces defining (i) a position measured at a given time for the object on the map, and (ii) accuracy information that defines a measure of accuracy of the measured position on the map;
• pairing successive traces in the sequence of traces;
• each pair of successive traces comprising a first trace and a second trace corresponding to two successive positions in time in the sequence of traces;
• defining an area of accuracy on the map using the accuracy information of the first trace and the second trace of each pair of successive traces;
• segmenting the map into tiles that includes the area of accuracy of the first trace and the second trace of each pair of successive traces;
• determining a set of road segments within the areas of accuracy of each pair of successive traces in the sequence of traces;
• for each segmented tile associated with each pair of successive traces in the sequence of traces, (i) computing a set of images representing a set of candidate paths of the object within the segmented tile;
• each candidate path in a set of candidate paths defining a different combination of road segments within the segmented tile;
• and (ii) computing, using a neural network, a set of local path scores representing a path score for each image in the set of images;
• defining a sequence of sets of local path scores with the set of local path scores computed for each segmented tile associated with each pair of successive traces in the sequence of traces;
• applying a neural graph model to the sequence of sets of local path scores to determine a most probable sequence of candidate paths that estimates the trajectory of the object on the map;
• defining a sequence of sets of images with the set of images computed for each segmented tile associated with each pair of successive traces in the sequence of traces;
• and
• outputting the images from the sequence of set of images defining the most probable sequence of candidate paths.

A method for estimating a trajectory of an object on a map, comprising:

• receiving a sequence of traces of the object on the map;
• pairing successive traces in the sequence of traces;
• segmenting the map into tiles that includes an area of accuracy of the first trace and the second trace of each pair of successive traces;
• for each segmented tile associated with each pair of successive traces in the sequence of traces, (i) computing a set of images representing a set of candidate paths of the object within the segmented tile;
• each candidate path in a set of candidate paths defining a different combination of road segments within the segmented tile;
• and (ii) computing, using a neural network, a set of local path scores representing a path score for each image in the set of images;
• defining a sequence of sets of local path scores with the set of local path scores computed for each segmented tile associated with each pair of successive traces in the sequence of traces;
• applying a neural graph model to the sequence of sets of local path scores to determine a most probable sequence of candidate paths that estimates the trajectory of the object on the map;
• defining a sequence of sets of images with the set of images computed for each segmented tile associated with each pair of successive traces in the sequence of traces;
• and
• outputting the images from the sequence of set of images defining the most probable sequence of candidate paths.