Probabilistic inference using weighted-integrals-and-sums-by-hashing for object tracking

摘要

A system for sensor fusion, typically for tracking objects near a vehicle, includes one or more sensors, a model component, and an inference component; a method is also disclosed. The model component is configured to calculate values in a joint-probabilistic graphical model based on the sensor data. The graphical model includes nodes corresponding to random variables and edges indicating correlations between the nodes; in a method this may include dividing the region near the vehicle into a grid and nodes may represent a binary occupancy of cells of the grid. The inference component is configured to detect and track obstacles near a vehicle based on the sensor data and the model, using a dimensionality reduction algorithm which most preferably is a weighted-integrals-and-sums-by-hashing (WISH) algorithm. Information about the location of the objects may be provided to an automatic driving or automatic assistance system, which may select a driving path using this information.