Close following detection using machine learning models
Described are embodiments for training and using a close following classifier. In the example embodiments, a system includes a backbone network configured to receive an image; and at least one prediction head communicatively coupled to the backbone network, the at least one prediction head configured to receive an output from the backbone network, wherein the at least one prediction head includes a classifier configured to classify the image as including a close-following event, the classifier receiving the output of the backbone network and a vehicle speed as inputs.
1 . A system comprising:
a backbone network configured to receive an image as an input, the backbone network comprising a machine learning model; and
at least one prediction head communicatively coupled to the backbone network, the at least one prediction head configured to
receive a vector output from the backbone network,
generate a combined input vector that concatenates the vector output from the backbone network and a vehicle speed recorded at a time associated with the image,
input the combined input vector into a classifier, and
obtain a classification of the image as representing a close-following event based on processing of the combined input vector by the classifier.
2 . The system of claim 1 , wherein the at least one prediction head includes a plurality of prediction heads and the plurality of prediction heads is trained using a joint loss function aggregating losses of each of the plurality of prediction heads.
3 . The system of claim 2 , wherein the plurality of prediction heads comprises a camera obstruction detection head, a lane detection head, an object detection head, and a distance estimation head.
4 . The system of claim 3 , wherein the classifier is configured to receive inputs from the lane detection head and the distance estimation head.
5 . The system of claim 4 , wherein the distance estimation head is configured to receive an input from the object detection head.
6 . The system of claim 3 , wherein the plurality of prediction heads further comprises an intermediate neural network, the intermediate neural network configured to process the output of the backbone network and transmit the output to the classifier.
7 . The system of claim 3 , further comprising an intermediate neural network configured to process the output of the backbone network and transmit the output to each of the plurality of prediction heads.
8 . The system of claim 2 , wherein the plurality of prediction heads comprises a camera obstruction detection head, a lane detection head, and an object bounding box, lane number and distance estimation head.
9 . The system of claim 8 , wherein the classifier is configured to receive inputs from the lane detection head and the object bounding box, lane number and distance estimation head.
10 . The system of claim 8 , wherein the object bounding box, lane number and distance estimation head is configured to receive an input from the lane detection head.
11 . A system comprising:
a backbone network configured to receive an image as an input, the backbone network comprising a machine learning model; and
a classifier configured to output a classification of the image as including a close-following event or not, the classifier receiving, as input, a combined input vector that comprises a concatenation of a vector output from the backbone network and a vehicle speed recorded at a time associated with the image, the classifier trained using a neural network, the neural network comprising at least one prediction head including the classifier, the at least one prediction head communicatively coupled to the backbone network based on processing of the combined input vector by the classifier.
12 . The system of claim 11 , further comprising an intermediate neural network communicatively coupled to the backbone network and the classifier, the intermediate neural network configured to process the output of the backbone network prior to transmitting the output to the classifier.
13 . The system of claim 11 , wherein the backbone network is configured to receive a video frame from a camera.
14 . The system of claim 13 , wherein the camera comprises a camera situated in a dash-mounted or windshield-mounted device.
15 . The system of claim 14 , wherein the backbone network and the classifier are executed on the dash-mounted or windshield-mounted device.
16 . A non-transitory computer-readable storage medium for tangibly storing computer program instructions capable of being executed by a computer processor, the computer program instructions comprising steps of:
receiving an image;
processing the image using a backbone network as an input, the backbone network comprising a machine learning model, an output of the backbone network comprising a set of features;
generating combined input vectors that each concatenate the set of features from the backbone network and a respective set of vehicle speeds recorded at a time associated with the set of features, wherein the at least one prediction head generates a prediction vector, and wherein the at least one prediction head includes a classifier configured to classify the image as including a close-following event based on processing of the combined input vectors; and
adjusting parameters of the backbone network and the at least one prediction head using a loss function.
17 . The non-transitory computer-readable storage medium of claim 16 , wherein the at least one prediction head further comprises an intermediate neural network, the intermediate neural network configured to process the output of the backbone network and transmit the output to the classifier.
18 . The non-transitory computer-readable storage medium of claim 16 , further comprising processing, using an intermediate neural network, the output of the backbone network, and transmitting the output to the at least one prediction head.
19 . The non-transitory computer-readable storage medium of claim 16 , wherein the at least one prediction head comprises a plurality of prediction heads including a camera obstruction detection head, a lane detection head, and an object bounding box, lane number and distance estimation head.
20 . The non-transitory computer-readable storage medium of claim 19 , wherein the loss function comprises a joint loss function aggregating individual loss functions of the plurality of prediction heads, each of the individual loss functions associated with a corresponding prediction head in the plurality of prediction heads.