Emergency vehicle movement prediction
|Title||Emergency vehicle movement prediction|
|Summary||Emergency vehicle movement prediction|
|Keywords||Smart city, cloud computing, Intelligent Transportation Systems, Artificial Intelligence, Machine Learning, Data mining|
|Prerequisites||Artificial Intelligence, Learning Systems|
|Supervisor||Cristofer Englund, Stefan Byttner|
Emergency vehicles (EVs) often require free-of-way and the sooner they can arrive at their destination the better are the chances to help the wounded. Today the EVs use sirens and lightning signals to alert the surrounding traffic and to create free-of-way. To further improve the ability for the surrounding traffic to receive the warning message that there is an EV approaching we assume that the EVs are able to share their location with the surrounding traffic when they are on a mission. The current data contains time and GPS coordinates of EVs. The sampling rate is between 30-20 seconds. This project should evaluate the data quality and explore what additional information could be added to improve the accuracy of a method that predicts the location of the EV at the next sampling time. Such a system can provide warnings to vehicles that are about to meet (or be overtaken) by an EV.
A simples method would give an indication in what area of the city the vehicle would appear in e.g. within a radius of the current location, whereas the more advanced, possibly a method based on machine learning, could learn, based on what road segment the vehicle is in, what time of the day it is and what the current speed of the vehicle is, to make a better prediction of where the EV will appear. And thus, a warning may be sent with wireless communication before vehicles will hear and see the siren or lightning.
From the available historical data also the destination is available for each trip. For SOS alarm however, it is a matter of patient security not to publically reveal any real addresses to where the EVs are heading. This project should also evaluate what SOS alarm would benefit in the prediction accuracy to also provide the destination (or some location very near the real destination).
The research questions that should be answered are: - What is the minimum amount of information you need to make a prediction about where the EV will be the next time sample? - Will the prediction be more accurate if you also provide the destination? - What are thre tradeoffs between giving false warnings and failing to provide a warning?
Possible methods that can be evaluated are: • Kalman filtering • Support vector machines • Neural networks • Random forest • Deep neural networks
The project could involve the following parts 1. Problem formulation and mapping of related work 2. Data extraction and mapping of GPS coordinates to map/road segments 3. Implementation of Algorithm 1 4. Implementation of Algorithm 2 5. Evaluation 6. Conclusions and suggestion on data modification, accuracy, sampling rate etc.