Counting Cars in Parking Lots

RESULTS
              

Project Objective

Large parking lots (say airports, universities, etc.) are studied to understand average parking availability with respect to time of the day and month. These reports are then used for future planning/expansion. Traditionally done using personnel on the ground, drones are now being used to do it within a fraction of the cost & time. Drone images captured are then analysed by human annotators which is an expensive process. Nanonets provides solutions to automate drone image analysis, again at a fraction of the cost & time.

Before

Traditionally, carrying out a parking study requires several people walking around the parking lot and manually counting the vehicles, multiples times in a day. For instance, a parking lot management company hired 6 personnel on ground to survey a parking lot consisting of 3,000 parking spaces. Data was collected on an hourly basis, 12 hours a day, for a period of 1 week. The data analysis and report generation typically took another week to complete. The data collection is prone to human error, partly due to the tedious and mundane nature of the task.

With the increase in availability of cheap camera mounted drones, it is not surprising that the use of drones for parking lot studies is on the rise.

After

By comparison, the same 3,000 spaces parking lot can be covered by a 15 minutes drone flight. The drone flies along a pre-planned route over the parking lot collecting overhead images at regular intervals. These images are then stitched together to form a huge orthomosaic image. Nanonets web API can then be used to get the count of cars in the orthomosaic image created each hour.

Each orthomosaic image is processed within seconds and the results are sent back to the server machine over the web. 1 human annotator then verifies the results of the API output and collates the results in his report.

Building a Nanonets Model

Building a custom Nanonets model for car counting took 3 days to build with training data of a 1,000 images. The easy-to-use APIs ensured that the integration and testing of Nanonet's custom built models was completed within a week.
This way, the use of drone and Nanonets APIs helped reduce the need for manual labour on ground and achieve a faster & more accurate analysis.

Nanonets Impact:

Humans involved for a parking lot study

Without Nanonets

8 persons

After integrating NanoNets API

2 persons

Cost of a typical parking lot study

Without NanoNets

$$$

After integrating NanoNets API

$