Inclement weather detection
The present application discloses a method, system, and computer system for detecting inclement weather driving conditions. The method includes obtaining an image captured by a camera mounted to a vehicle, determining a classification for road and weather conditions using a condition prediction model to analyze the image, in response to determining that the classification for road and weather conditions matches a particular predefined road and weather classification, determining an active measure associated with the particular predefined road and weather classification, and causing the active measure to be performed.
1 . A system, comprising:
a memory;
one or more processors configured to:
obtain an image captured by a camera mounted to a first vehicle;
obtain weather data for a location at which the image was captured, wherein the weather data is obtained from map data or based at least in part on information obtained from a third-party weather service;
determine a classification for road and weather conditions using a condition prediction model to analyze the image and the weather data, wherein the condition prediction model generates the classification based on a fused representation of the image and the weather data; and
in response to determining that the classification for road and weather conditions matches a particular predefined road and weather classification,
determine an active measure associated with the particular predefined road and weather classification; and
cause the active measure to be performed.
2 . The system of claim 1 , wherein the active measure comprises providing an alert to a driver of the first vehicle.
3 . The system of claim 2 , wherein the alert includes a recommendation for the driver to change a speed of the first vehicle.
4 . The system of claim 1 , wherein the active measure includes re-routing the first vehicle.
5 . The system of claim 1 , wherein the active measure includes providing an alert to a fleet manager of a fleet associated with the first vehicle.
6 . The system of claim 5 , wherein the active measure further includes providing one or more recommended alternative routes for one or more managed vehicles of the fleet.
7 . The system of claim 1 , wherein the active measure is determined based at least in part on querying a mapping of types of road and weather classifications to active measures.
8 . The system of claim 1 , wherein the condition prediction model comprises a machine learning model.
9 . The system of claim 8 , wherein the condition prediction model is trained using a deep neural network.
10 . The system of claim 1 , wherein determining the classification for road and weather conditions comprises:
determining a confidence score; and
determining the classification for the road and weather conditions based at least in part on the confidence score.
11 . The system of claim 10 , wherein determining the classification for the road and weather conditions based at least in part on the confidence score comprises:
determining whether the confidence score exceeds a predefined confidence threshold; and
in response to determining that the confidence score exceeds the predefined confidence threshold, deeming the road and weather conditions to be the particular predefined road and weather classification.
12 . The system of claim 11 , wherein determining the active measure comprises:
obtaining a vehicle speed; and
determining the active measure based at least in part on the vehicle speed.
13 . The system of claim 12 , wherein the active measure includes a recommendation to change the vehicle speed based on a percentage of the vehicle speed.
14 . The system of claim 1 , wherein the condition prediction model generates a prediction according to an N-class weather classification, N being a positive integer.
15 . The system of claim 14 , wherein:
N is equal to 3; and
three classifications of the N-class weather classification include:
(i) a clear road condition, and a clear or snowy weather condition;
(ii) a wet road condition, and a rainy or snowy weather condition; and
(iii) (a) any road condition and a foggy or icy weather condition; or
(b) an icy or snowy road condition and a snowy weather condition.
16 . The system of claim 1 , wherein the condition prediction model is trained based on a sampling of a set of images captured by one or more managed vehicles.
17 . The system of claim 16 , wherein the sampling of the set of images is generated based on predicted classifications for images comprised in the set of images.
18 . The system of claim 17 , wherein the sampling of the set of images for the particular predefined road and weather classifications comprises selected image samples for which corresponding predictions correspond to the particular predefined road and weather classification by having associated confidence scores that exceed a threshold.
19 . The system of claim 1 , wherein:
the image is an image for a particular road;
the active measure comprises rerouting a second vehicle in response to determining, based at least in part on a planned route for the second vehicle, that the second vehicle is expected to travel along the particular road within a threshold distance of the location at which the image was captured and within a threshold period of time from when the image was captured.
20 . The system of claim 1 , wherein the active measure is caused to be performed only in response to determining that a confidence score associated with the classification exceeds a predefined threshold corresponding to that active measure.
21 . The system of claim 1 , wherein the condition prediction model is retrained based on a sampling of images selected according to predicted classifications having confidence scores that exceed a retraining threshold.
22 . The system of claim 1 , wherein in response to determining the classification, the active measure further comprises transmitting an alert or rerouting instruction to one or more additional vehicles of a fleet managed with the first vehicle.
23 . The system of claim 1 , wherein in response to determining the classification, the active measure further comprises transmitting an alert or rerouting instruction to one or more additional vehicles of a fleet managed with the first vehicle.
24 . A method, comprising:
obtaining an image captured by a camera mounted to a first vehicle;
obtain weather data for a location at which the image was captured, wherein the weather data is obtained from map data or based at least in part on information obtained from a third-party weather service;
determining a classification for road and weather conditions using a condition prediction model to analyze the image and the weather data, wherein the condition prediction model generates the classification based on a fused representation of the image and the weather data; and
in response to determining that the classification for road and weather conditions matches a particular predefined road and weather classification,
determining an active measure associated with the particular predefined road and weather classification; and
causing the active measure to be performed.
25 . A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:
obtaining an image captured by a camera mounted to a first vehicle;
obtain weather data for a location at which the image was captured, wherein the weather data is obtained from map data or based at least in part on information obtained from a third-party weather service;
determining a classification for road and weather conditions using a condition prediction model to analyze the image and the weather data, wherein the condition prediction model generates the classification based on a fused representation of the image and the weather data; and
in response to determining that the classification for road and weather conditions matches a particular predefined road and weather classification,
determining an active measure associated with the particular predefined road and weather classification; and
causing the active measure to be performed.