Alvelor is an acronym for ALgorithmic VEhicular LOad Recognition and is an open source project to use traffic cameras to extract information about the volume of vehicles on the road. It uses image recognition algorithms to analyse traffic camera feeds and extract information about the number and type of vehicles that are visible. With this information it is possible to generate datasets about traffic volume throughout the day, week month and year.
Solution
Currently the process includes two steps: first step is to scrape and store images from the cameras every minute. Second step is to process that with machine learning and store the resulting data in a structured format. We have built the solution on AWS technologies but it can be deployed on any IaaS or on premise infrastructure.
Open tasks
We have a number of tasks open that would improve the solution
- Optimize image quality – some images are not very good quality which impacts the model performance. Especially images at night time degrade the quality of the model. We need to look at how images can be improved to optimize the model output.
- Optimize model – the model can be optimized and should be tuned. There are some false positives and false negatives that could be eliminated if the model was optimized.
- Analyze failed images – public traffic camera feeds are provided on an as is basis and they are frequently down or unavailable. We need to create a process to detect when a camera is down or no image was stored.
- Detect camera feed changes – We have noticed that sometimes that traffic cameras turn or the feed changes altogether. This affects the counts since angle, position and direction all impacts how the image is processed. We need to be able to detect when an image differs from a canonical image.
How to participate
Everyone is encouraged to participate in whatever capacity they can. Contact us through the contact form to understand more and to find out how to contribute
alvelor 