Traffic light control based on traffic pattern prediction
Disclosed herein are systems and methods for controlling traffic lights according to predicted traffic patterns, comprising receiving one or more image sequence comprising a plurality of images captured by one or more imaging sensor deployed to monitor vehicle traffic in one or more intersection in which traffic light(s) is deployed to control traffic flow, generating a traffic dataset descriptive of time series movement of all vehicles tracked in the image sequence(s), applying a first trained machine learning model to map, based on the traffic dataset, a traffic pattern of the tracked vehicles to one or more of a plurality of learned traffic patterns, applying a second trained machine learning model to predict one or more subsequent traffic patterns based on the mapped traffic pattern, and generating instructions for controlling the traffic light(s) according to the predicted subsequent traffic pattern(s).
1 . A computer implemented method of controlling traffic lights according to predicted traffic patterns, comprising:
receiving at least one image sequence comprising a plurality of images captured by at least one imaging sensor deployed to monitor vehicle traffic in at least one intersection, at least one traffic light is deployed in the at least one intersection to control traffic flow;
generating a plurality of vehicle time-series movement patterns, each descriptive of movement over time of a respective vehicle tracked in the at least one image sequence, and a plurality of lane time-series movement patterns, each descriptive of movement over time of vehicles in a respective lane of the at least one intersection;
generating a traffic dataset based on the plurality of vehicle time-series movement patterns and the plurality of lane time-series movement patterns, the traffic dataset descriptive of time series movement of all vehicles tracked in the at least one image sequence;
applying a first trained machine learning model to map, based on the traffic dataset, a traffic pattern of the tracked vehicles to at least one of a plurality of learned traffic patterns, thereby creating a consecutive mapping sequence comprising consecutive traffic patterns mapped for traffic flowing through the at least one intersection over time;
applying a second trained machine learning model to predict at least one subsequent traffic pattern based on the consecutive mapping sequence; and
generating instructions for controlling the at least one traffic light according to the at least one predicted subsequent traffic flow.
2 . The computer implemented method of claim 1 , further comprising generating the instructions for controlling the at least one traffic light according to a control plan selected based on a simulation of a plurality of control plans applied to control the at least one traffic light for controlling a flow of vehicles defined by the at least one predicted subsequent traffic flow.
3 . The computer implemented method of claim 2 , wherein the simulation is directed to predict a flow of vehicles through the at least one intersection where the selected control plan is estimated to induce optimal flow expressed by a reduced time for the vehicles to pass the at least one intersection.
4 . The computer implemented method of claim 1 , wherein the traffic dataset comprises at least one of: at least one vehicle parameter of each tracked vehicle and at least one lane parameter of each lane in the at least one intersection, the at least one vehicle parameter and the at least one lane parameter are identified based on analysis of the at least one image sequence.
5 . The computer implemented method of claim 4 , wherein the at least one vehicle parameter is a member of a group consisting of: a vehicle type, a lane, a position in the lane, a position in a queue in the lane, a location, a relative location with respect to at least one another vehicle, a type of adjacent vehicles, a speed, an acceleration, a wait time at the at least one intersection, a distance form a stop line of the at least one intersection, and an overall tracking time.
6 . The computer implemented method of claim 4 , wherein the at least one lane parameter is a member of a group consisting of: a number of vehicles in the lane, a type of vehicles in the lane, an order of the vehicles in of a queue in the lane, a length of the queue and a lane crossing time duration.
7 . The computer implemented method of claim 4 , wherein the analysis further comprises filtering out at least one object unrelated to tracked vehicles detected in the at least one image sequence.
8 . The computer implemented method of claim 4 , wherein the analysis further comprises applying at least one trained model to track at least one partially visible vehicle in the at least one image sequence, the at least one partially visible vehicle is at least partially invisible in at least one of the plurality of images.
9 . The computer implemented method of claim 1 , wherein the first machine learning model is trained using a plurality of traffic datasets generated based on a plurality of previously captured image sequences of the at least one intersection.
10 . The computer implemented method of claim 1 , wherein the first machine learning model is trained in at least one unsupervised training session to map the plurality of traffic patterns of vehicles detected at the at least one intersection to a plurality of respective clusters.
11 . The computer implemented method of claim 1 , wherein the first machine learning model is further trained post-deployment using a plurality of traffic datasets generated based on a plurality of image sequences captured after the deployment.
12 . The computer implemented method of claim 1 , wherein the second machine learning model is trained in at least one supervised training session using a plurality of consecutive mapping sequences of a plurality of traffic datasets generated based on a plurality of previously captured image sequences of the at least one intersection.
13 . The computer implemented method of claim 1 , wherein the second machine learning model is further trained post-deployment using a plurality of consecutive mapping sequences of a plurality of traffic datasets generated based on a plurality of image sequences of the at least one intersection captured after the deployment.
14 . The computer implemented method of claim 1 , wherein at least part of the process to control the at least one traffic light is executed by an edge node deployed at the at least one intersection which is functionally coupled to the at least one imaging sensor.
15 . The computer implemented method of claim 1 , wherein at least part of the process to control the at least one traffic light is executed by a remote server which is communicatively coupled to the at least one imaging sensor via at least one network.
16 . A system for controlling traffic lights according to predicted traffic patterns, comprising:
at least one processor executing a code, the code comprising:
code instructions to receive at least one image sequence comprising a plurality of images captured by at least one imaging sensor deployed to monitor vehicle traffic in at least one intersection, at least one traffic light is deployed in the at least one intersection to control traffic flow;
code instructions to generate a plurality of vehicle time-series movement patterns, each descriptive of movement over time of a respective vehicle tracked in the at least one image sequence, and a plurality of lane time-series movement patterns, each descriptive of movement over time of vehicles in a respective lane of the at least one intersection;
code instructions to generate a traffic dataset based on the plurality of vehicle time-series movement patterns and the plurality of lane time-series movement patterns, the traffic dataset descriptive of time series movement of all vehicles tracked in the at least one image sequence;
code instructions to apply a first trained machine learning model to map, based on the traffic dataset, a traffic pattern of the tracked vehicles to at least one of a plurality of learned traffic patterns;, thereby creating a consecutive mapping sequence comprising consecutive traffic patterns mapped for traffic flowing through the at least one intersection over time;
code instructions to apply a second trained machine learning model to predict at least one subsequent traffic pattern based on the consecutive mapping sequence; and
code instructions to generate instructions for controlling the at least one traffic light according to the at least one predicted subsequent traffic flow.
17 . A computer program product for controlling traffic lights according to predicted future traffic flow, comprising a non-transitory medium storing thereon computer program instructions which, when executed by at least one hardware processor, cause the at least one hardware processor to:
receive at least one image sequence comprising a plurality of images captured by at least one imaging sensor deployed to monitor vehicle traffic in at least one intersection, at least one traffic light is deployed in the at least one intersection to control traffic flow;
generate a plurality of vehicle time-series movement patterns, each descriptive of movement over time of a respective vehicle tracked in the at least one image sequence, and a plurality of lane time-series movement patterns, each descriptive of movement over time of vehicles in a respective lane of the at least one intersection;
generate a traffic dataset based on the plurality of vehicle time-series movement patterns and the plurality of lane time-series movement patterns, the traffic dataset descriptive of time series movement of all vehicles tracked in the at least one image sequence;
apply a first trained machine learning model to map, based on the traffic dataset, a traffic pattern of the tracked vehicles to at least one of a plurality of learned traffic patterns, thereby creating a consecutive mapping sequence comprising consecutive traffic patterns mapped for traffic flowing through the at least one intersection over time;
apply a second trained machine learning model to predict at least one subsequent traffic pattern based on the consecutive mapping sequence; and
generate instructions for controlling the at least one traffic light according to the at least one predicted subsequent traffic flow.