Road surface characterization using pose observations of adjacent vehicles

A system comprising one or more processing devices and one or more memory devices coupled to the one or more processing devices, the one or more memory devices storing executable code effective to cause the one or more processing devices to:

• receive training data entries, each entry including a video segment annotated with a vehicle pose and a hazard type;
• train a first deep neural network (DNN) to output vehicle pose estimates according to the training data entries;
• and
• train a second DNN to classify a hazard according to the training data entries and the vehicle pose estimates,
• wherein the executable code is further effective to cause the one or more processing devices to train the second DNN using the training data entries to identify a location of the hazard according to the vehicle pose estimates, and derivatives of the vehicle pose estimates.

A method comprising, by a computer system:

• receiving training data entries, each entry including a video segment captured by a camera mounted to a first vehicle and annotated with a vehicle pose of a second vehicle visible in the video segment and a hazard type being encountered by the second vehicle visible in the video segment;
• training a first deep neural network (DNN) using the training data entries to output a vehicle pose estimate for a third vehicle visible in video data captured by a camera mounted to a fourth vehicle;
• and
• training a second DNN using the training data entries to classify a hazard encountered by the third vehicle according to the vehicle pose estimate for the third vehicle.