Parking Lot Occupancy Database
Database for evaluating the ParkingNet system that estimates the parking lot occupancy in complex scenarios. The system used a specific neural network architecture that can be trained with a database with a reduced annotation scheme, reducing costs and making faster the system deployment. The database contains images of a complex parking scenario (day and night, perspective occlusions, background occlusions, etc.) using the following folder structure:
Images: folder containing a low frame rate image sequence of a parking lot.
groundtruth.txt: ground truth information about the occupancy in the parking lot.
Every line contains the occupancy ground truth of an image according to the following format: [image_name 21-bit-vector]. See an example below.
21-bit-vector: every component indicates if one parking space is available (1) or not (0). The relationship between parking spaces and the visual images is indicated in the figure below. There is no need for explicit image region information of the parking spaces, speeding up the creation of the database annotation.
Validation: same structure than Training folder.
Test: same structure than Training folder.
Example of ground truth information
Example of parking lot image (100GOPRO-GOPR0650.JPG) with indications for relating to ground truth and visual information