Sensor fusion and deep learning

A method, comprising:

• performing, by one or more computer systems:
• preprocessing a plurality of sensor data representations of an environment, generated by a plurality of separate sensor devices, to combine the plurality of sensor data representations into a fused sensor data representation of the environment;
• after combining the sensor data representations, detecting a feature, associated with at least one object, in the fused sensor data representation of the environment;
• identifying one or more object classifications for the feature as detected in the fused sensor data representation;
• detect the feature associated with the at least one object in respective ones of the plurality of sensor data representations of the environment;
• determine corresponding one or more classification confidence values for the one or more object classifications based on the feature as detected in the respective ones of the plurality of sensor data representations;
• and
• classifying the at least one object according to the one or more object classifications and the one or more confidence values.

A non-transitory, computer-readable medium storing a program of instructions which, when executed by at least one computer system, causes the at least one computer system to:

• preprocess a plurality of sensor data representations of an environment, generated by a plurality of separate sensor devices, to combine the plurality of sensor data representations into a fused sensor data representation of the environment;
• and
• after combining the sensor data representations, detect a feature, associated with at least one object, in the fused sensor data representation of the environment;
• identify one or more object classifications for the feature as detected in the fused sensor data representation;
• detect the feature associated with the at least one object in respective ones of the plurality of sensor data representations of the environment;
• determine corresponding one or more classification confidence values for the one or more object classifications based on the feature as detected in the respective ones of the plurality of sensor data representations;
• and
• classify the at least one object according to the one or more object classifications and the one or more classification confidence values.