Managing traffic and vehicles is a compex task. Nowadays, governments expend huge amounts of resources to create optimized traffic of vehicles managing. One of the most important facets if this is the parking managing. Some cities have vacant parking spots counters for the streets and parking lots. However, they are very approximate and do not tell where exactly are the free spots.
In this project we propose PARKOUR, a computer vision-based smart parking lot manager. Our system is able to automatically detect the position of the vehicles so it knows which spots are vacant or taken.
The proposed pipeline is split in two stages. First, a human operator is in charge of labelling the parking spots so the system knows where they are located. And then, PARKOUR takes the images from the camera and feeds them to a region convolutional neural network that provides the position of the vehicles in the image. With the position of the vehicles we can detect which spots are taken. After all vehicles have been processed, the vacant and taken sports are reported by PARKOUR. We also report if there are vehicles parked outside the permitted parking space.We also extract the position of individuals so the system can detect possible suspicious behaviour. For instance, if a person is roaming around the parking lot too much time. As a requirement, an outdoor camera pointing to the parking lot is needed.
The benefits of our system over traditional parking managing systems is that there is no need to install gates with ticket systems to compute how many vehicles are there in the parking lot. In addition, our approach is able to provide the list of taken and vacant parking spots, which is useful for the user so it minimizes the time they take to find a free spot.
The University of Alicante (Spain) has granted us permission to develop and actually test our approach in one of their parking lots.
Finally is it worth mentioning that a substantial part of this project will be carried out using OpenCV. For instance, it will be used for the labeling, bounding boxes management and displaying of information.
Comments