Traffic Light Control Using Local Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current traffic light control systems lack the ability to dynamically adjust to local conditions and prioritize safety for non-vehicular travelers, leading to inefficiencies in traffic flow and increased congestion, which can result in longer travel times and frustration for drivers.
Innovation Solution
A traffic light control system that utilizes local machine learning models to analyze camera and sensor data from intersections, generating real-time control instructions that can override global instructions to prioritize pedestrian and cyclist safety, while also optimizing traffic flow by coordinating phases across multiple intersections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traffic light phases are coordinated to create green waves for vehicles, then traffic flow efficiency is improved, but pedestrian and cyclist safety is compromised
Solution Approach 1:
The system dynamically adjusts traffic light phases based on real-time detection of vulnerable road users. When pedestrians or cyclists are detected near crosswalks, the system extends green light phases to ensure their safety, while maintaining green wave coordination for vehicles when no vulnerable users are present. This dynamic adaptation resolves the contradiction by making the system flexible rather than fixed.
Solution Approach 2:
The system uses machine learning models to continuously monitor and detect vulnerable road users, providing feedback to the traffic light control system. This feedback loop enables the system to adjust phases in real-time, extending green lights when pedestrians or cyclists are detected and maintaining efficient green waves when the road is clear, thus balancing safety and efficiency.
2Reliability
If traffic light phases are extended to protect pedestrians, then pedestrian safety is improved, but vehicle travel time increases
Solution Approach 1:
The system applies pedestrian protection measures partially and selectively rather than continuously. Green light phases are extended only when and where vulnerable road users are detected, not at all times or all locations. This partial application of safety measures minimizes impact on vehicle travel time while still providing necessary protection when needed.
Solution Approach 2:
The system implements different traffic control strategies at different locations and times based on local conditions. At intersections where pedestrians or cyclists are detected, the system extends green phases locally. At other intersections or times when no vulnerable users are present, the system maintains standard timing, thus protecting pedestrians where necessary without universally increasing vehicle travel time.
3Reliability
If local machine learning models are used to detect vulnerable road users, then safety detection capability is improved, but system complexity increases
Solution Approach 1:
The system uses machine learning models as intermediary components between the raw sensor data and the traffic light control decisions. These models process sensor inputs to detect vulnerable road users and translate them into actionable information for the control system, simplifying the overall architecture while improving detection capability.
Solution Approach 2:
The machine learning models operate autonomously to detect and classify vulnerable road users from sensor data without requiring constant human intervention or complex centralized processing. The local models self-manage the detection task, reducing the burden on the overall system architecture and enabling distributed intelligence at each intersection.
Data Source
AI summary
In some implementations, a method includes receiving, by one or more processing devices configured to control a traffic signal at an intersection of roads, camera data providing images of the intersection, the processing devices being located proximate to the intersection, using one or more local machine learning models to identify objects at the intersection and paths of the objects based on the camera data, providing traffic data generated from outputs of the one or more local machine learning models to a remote traffic planning system over a network, receiving, from the remote traffic planning system, a remote instruction for the traffic signal determined using one or more remote machine learning models, and providing a control instruction to the traffic signal at the intersection that is determined based on (i) the remote instruction from the remote traffic planning system, and (ii) a local instruction generated by the processing devices.