Problem
Previous road management systems ran independent of traffic and car information. This results in many inconveniences:
- Lost productivity and wasted time: Everyday traffic congestion promotes loss of productivity for companies. In many countries, employees spend an excessive amount of time in traffic when they could be working on their jobs.
- High number of traffic accidents: Inefficient use of traffic lights increases the risk of car accidents.
- High number of pedestrian accidents: a study suggests that walking in traffic situations is 10 times more dangerous than travelling as a passenger by car. Moreover, this study also suggests that 15% of total people killed in European roads are pedestrians.
- High environmental cost and increased energy spent waiting in traffic: The stop-start driving and long time waiting in traffic is inefficient and very polluting to cities.
Addressing these inconveniences by using a computer vision application (real time smart traffic lights) helps drivers, pedestrians, municipalities and even businesses (gains in productive time).
Solution
The solution proposed encompasses two main elements:
- Two cameras located on one post of a traffic light: one oriented to capture pedestrians, and the other to capture car traffic
- A machine learning algorithm that is capable of:
- Recognizing “new elements” in every frame, the camera captures – be it pedestrians or cars
- Predicting (i) the trajectory of these elements using AI, (ii) the intensity of future traffic
- Making real-time decisions on the “position” of traffic lights for both cars (green, yellow, red) and pedestrians (white, red).
Effectiveness and commercial promise
For pedestrians, previous applications either didn’t take input from pedestrian traffic or used a manual push-button for input. The proposed solution is definitely much more attractive to pedestrians: it does not require any additional action from their end and still manages to improve their outcomes.
For car traffic, previous smart traffic systems used a system called an inductive loop detector, that is embedded under the street. The benefit of the new solution compared to this lies in the following:
- Recent studies have shown loop inaccuracies of up to 20%, when compared with actual footage. This lack of reliability, especially in extreme congestion where these solutions are most needed, hinders extreme optimization of traffic
- Detection of faulty loops is very difficult, and maintaining them often requires blocking roads, which goes against the objective of easing traffic
In terms of outcomes, the prototypes for this technology have started being tested and have shown promising results especially in smooth light-change conditions and when few objects are present.
Anticipated competition
Companies that have access to big sources of data have an advantage to provide solutions using machine learning. Among those companies we could give the example of Google and Microsoft that are trying to revolutionize some industries by developing advanced analytics for their customers. In terms of traffic management solution these companies could create partnerships with the government and municipalities to develop integral solutions to enhance the traffic management. Currently they have showed some efforts in the field:
Google: The company is already involved in the traffic management field with his applications Google Maps and Waze. Those could help them have a important source of information to continue developing solutions. Recently they have backed up a venture, through google ventures, called Urban engines that is working in aggregating traffic data to create predictive models that will give users the ability to improve routes management and avoid congestion based on what time of day it is and what’s happening in the area.
Microsoft: Microsoft has azure, his cloud analytics tool. With it they are helping companies develop specific tools to improve their businesses through analytics. In terms of traffic management they are involved in developing sustainable cities through CityNext, an initiative that they are sharing with various companies, such as Cubic, to solve transportation issues.
Alterations proposals
Apart from the current, and above mentioned, applications that computer vision is being used for in traffic lights, we propose the following different uses from which society can benefit from:
- Response to emergency cars (e.g. ambulances, fire trucks). Many more lives could be saved if ambulances and fire trucks arrived on time at the emergency point by having traffic lights change in their favor.
- In many countries, specially high-crime countries in Mexico and LatAm, many robberies and kidnaps occur during red lights. Some countries have allowed drivers to run pass a red light after midnight for security reasons. However, this increases car accidents. Therefore, crime prevention is another great use for this technology.
- With the use of the cameras and machine learning algorithm, it will be possible possible to have a more accurate speed management in which the speeding car can be recognized through the cameras.
Sources:
Usefulness of image processing in urban traffic control, Boillot (https://ac.els-cdn.com/S1474667017438730/1-s2.0-S1474667017438730-main.pdf?_tid=efc0cf01-ae0b-4c44-831c-4295172c54b0&acdnat=1522637076_32242720358f549ac1c0a6a27df47e31)
Computer vision application: Real time smart traffic light (https://pdfs.semanticscholar.org/d1c3/bfc4e8ff2861137da2af817e4fbe709339da.pdf)
Traffic management startup backed by google: Urban engines
A New Smart Technology will Help Cities Drastically Reduce their Traffic Congestion
Microsoft CityNext
(https://enterprise.microsoft.com/en-us/industries/citynext/sustainable-cities/transport/)
Team:
Francisco Galvez
Stephanie Saade
Marisol Perez-Chow
Luca Ferrara
Caitlyn Grudzinski
Wing Kiu Szeto