Overview / Usage

V-Spark is a research project that aims to overcome some of the limitations of the current smart parking sensor technology, which uses an array of sensors to count the occupancy of vehicles in parking spaces. The goal for the V-Spark project is to use the Intel RealSense 3D camera technology combined with Intel OpenVino to act as a "Virtual Smart Parking Sensor" (V-Spark).

V-Spark will integrate into an existing smart parking counting and reporting network infrastructure (LoRaWAN, NB-IoT, Sigfox, etc) and act as a "virtual sensor" which will allow it to count a minimum of 20 parking spaces at one time. The goal is to have V-Spark to count up to 100 parking spaces with 99.5% accuracy in all weather conditions and lighting conditions.

Methodology / Approach

The Problem:

In 2017, USA Today reports that drivers waste over 17 hours per year looking for available parking: https://eu.usatoday.com/story/money/2017/07/12/parking-pain-causes-financial-and-personal-strain/467637001/

This of course, is the reason why "smart parking" is one of the most popular solutions requested by drivers in order to transform an ordinary city into a "smart city". The current state-of-the-art technology used for vehicle detection uses a single, weather-proof sensor that is installed on the surface and detects an individual vehicle through radio waves or induction. However, there are several drawbacks to this approach:

1. The single-sensor-single-vehicle approach has scalability problems. In large cities such as London, it has been estimated that there are over 6 million parking spaces! https://londontransportdata.wordpress.com/category/subject/parking/ This makes the cost of deploying a comprehensive, city-wide solution to be prohibitively expensive (simply on a hardware basis alone)
2. Each sensor needs to be physically installed into a hard surface (road, pavement, concrete, etc) in order to sense the vehicles. This labor cost adds to the total cost of ownership of a smart parking solution due to the involvement of a maintenance crew and drilling for each sensor. Additionally, for cities that have cobblestone streets (Philadelphia, Seattle, London, Paris, Copenhagen, etc.) drilling into the stones of the street is highly discouraged.
3. Sensors that are inserted BELOW the ground level have a potential to not detect a vehicle properly if a puddle of water is over the sensor
4. Sensors that are inserted SLIGHTLY ABOVE the ground level cannot be used in climates where snow plows are used to clear the snow from the street
5. Each sensor needs its own battery in order to communicate the presence (or absence) of a vehicle
6. When a battery in the sensor dies (or when the sensor fails) the municipality needs to weigh the cost/benefit to repair/replace a single sensor one at a time
7. It is relatively easy to know when the battery has died, but it is difficult to ascertain if the sensor has failed (false positives or false negatives)

Technologies Used

The Solution:

The goal of the V-Spark project is to create a "virtual sensor" that uses computer vision and pre-trained AI models to detect when a vehicle is occupying a parking space. V-Spark is an IoT Edge device that uses the following technologies:

• Intel RealSense 3D camera to capture the parking surface vehicle in 3D and low-lighting conditions (ideally mounted on a light pole)
• Intel CVAT (Computer Vision Annotation Tool) to annotate the 3D images of the vehicles to train the model
• Intel AI DevCloud/Xeon for model training and export
• Intel OpenVINO Toolkit with Movidius NCS for vehicle detection on 3D images