Fusion of multi-sensor information with operator-learned behavior for automatic and efficient recognition of objects and control of remote vehicles

摘要

Systems and methods are described for remotely controlled vehicles that hierarchically transform sensor-detected information into levels of increasing fidelity for highly efficient machine recognition of attributes and features of detected objects/events. Fusion of the feature space representations of the various levels allows correlation with the operator's attention to the respective objects for automated processing/controlling of the unmanned vehicle. High efficiencies in object/event recognition are attained with reduced memory and processor requirements, enabling near autonomous or fully autonomous operation of the vehicle.

主权项

1. A computerized method for machine recognition of objects or events from a remotely controlled, unmanned vehicle, comprising:
electronically receiving a signal from a plurality of sensors deployed on the vehicle; determining signal-specific components for partitioning each signal; building sets of independent levels for each received signal representing hierarchal information from the signals, each set containing levels spanning from a lowest information level (i=0), where i is an integer, to a highest information level (i=n), where n is a maximum integer; evaluating each level (i) within its set with its associated signal-specific components; building sets of vector signatures associating each level (i) with its evaluated associated signal-specific components; correcting, via a remote human operator of the vehicle, an accuracy of the vector signatures; combining the sets of vector signatures from each signal to form fused signature vectors; mapping natural language semantics to match the fused signature vectors, beginning at lower levels and increasing until a match is found; assigning unique matched semantics to each level(i); pairing syntaxes with the unique semantics and saving the associated fused signature vectors; and eliminating fused signature vectors that result in non-deterministic matches,
wherein resulting sets of fused signature vectors and their syntaxed unique semantics define a transformation of information from the signals into machine generated descriptions of objects or events; andwherein the building sets of independent levels is performed according to:
level (i=n) if it contains 2n×2n bits of information; and signal information may be strictly converted from level n to level n−1 according to:⋃i=02n-1-1⁢⋃j=02n-1-1⁢ai,j=(1,δi,j>threshold0,otherwise;)where,⁢δi,j=a2⁢i+1,2⁢j+1+a2⁢i+2,2⁢j+1+a2⁢i+1,2⁢j+2+a2⁢i+2,2⁢j+24, where variable “a” is a subscripted pixel.