DETERMINING PARKING SPACE OCCUPATION IN COMPUTER VISION SYSTEMS
Authors:
1.
Akademiya nauk Respubliki Tatarstan
GBU «Bezopasnost' dorozhnogo dvizheniya»
GBU «Bezopasnost' dorozhnogo dvizheniya»
2.
Kazan National Research Technical University named after A.N. Tupolev
3.
Kazan State Energy University
4.
Kazan National Research Technical University named after A.N. Tupolev
5.
KNITU-KAI
6.
GBU «Bezopasnost' dorozhnogo dvizheniya»
Type:
Article
eLocation:
Status:
Published
Received:
05.08.2025
Accepted:
07.08.2025
Published:
07.08.2025
Language:
Russian
Keywords:
OBJECT RECOGNITION, OBJECT INTERSECTION METRICS, CONVEX HULL INTERSECTION, INTERSECTION OVER UNION, PARKING SPACE, SMART PARKING
Abstract (English):
The article is devoted to the description and comparison of metrics of the degree of intersection of objects in determining the occupancy of parking spaces. The choice of metric is crucial, since the accuracy of determining the fact of occupancy or unoccupancy of a parking space depends on it. The features of the main metrics of the degree of intersection of objects are considered: the ratio of intersection of areas and the union of areas (Intersection over Union), the ratio of intersection of areas and the area of an object (Intersection over Object Area), the ratio of intersection of areas and the area of a mask (Intersection over Mask Area), the intersection of minimum bounding rectangles (MBR Intersection), the distance between centroids (Centroid Distance) and the intersection of convex hulls (Convex Hull Intersection). A testing procedure has been developed to evaluate the accuracy, efficiency and running time of algorithms based on each of these metrics. It includes obtaining a two-hour video clip from a parking surveillance camera, marking the contours of parking spaces for the camera angle used, manually marking the occupancy or vacancy of each parking space in the video sequence, recognizing vehicles in each frame using the YOLO 11n model, determining the occupancy or vacancy of all parking spaces in the frames of the video sequence using each of the described metrics, calculating the average processing time of one frame for each metric, calculating the accuracy, precision, recall and F1-score values to assess the accuracy of each of the metrics of the degree of intersection of objects. The process of testing and comparing the metrics of the degree of intersection of objects was carried out on a computer with an AMD Ryzen 5 5000U processor and 16 GB of RAM. The results of the testing and comparison of the metrics of the degree of intersection of objects are presented. The Intersection over Object Area and Convex Hull Intersection metrics showed the best results. The Centroid Distance and MBR Intersection metrics showed lower accuracy, since they do not take into account the shape of objects and can produce false results. The worst metrics were Intersection over Mask Area and Intersection over Union, both in accuracy and in the average frame processing time. The analysis showed that the Intersection over Object Area and Convex Hull Intersection metrics solve the task most accurately. This determines the feasibility of using them for monitoring urban parking spaces.
The article is devoted to the description and comparison of metrics of the degree of intersection of objects in determining the occupancy of parking spaces. The choice of metric is crucial, since the accuracy of determining the fact of occupancy or unoccupancy of a parking space depends on it. The features of the main metrics of the degree of intersection of objects are considered: the ratio of intersection of areas and the union of areas (Intersection over Union), the ratio of intersection of areas and the area of an object (Intersection over Object Area), the ratio of intersection of areas and the area of a mask (Intersection over Mask Area), the intersection of minimum bounding rectangles (MBR Intersection), the distance between centroids (Centroid Distance) and the intersection of convex hulls (Convex Hull Intersection). A testing procedure has been developed to evaluate the accuracy, efficiency and running time of algorithms based on each of these metrics. It includes obtaining a two-hour video clip from a parking surveillance camera, marking the contours of parking spaces for the camera angle used, manually marking the occupancy or vacancy of each parking space in the video sequence, recognizing vehicles in each frame using the YOLO 11n model, determining the occupancy or vacancy of all parking spaces in the frames of the video sequence using each of the described metrics, calculating the average processing time of one frame for each metric, calculating the accuracy, precision, recall and F1-score values to assess the accuracy of each of the metrics of the degree of intersection of objects. The process of testing and comparing the metrics of the degree of intersection of objects was carried out on a computer with an AMD Ryzen 5 5000U processor and 16 GB of RAM. The results of the testing and comparison of the metrics of the degree of intersection of objects are presented. The Intersection over Object Area and Convex Hull Intersection metrics showed the best results. The Centroid Distance and MBR Intersection metrics showed lower accuracy, since they do not take into account the shape of objects and can produce false results. The worst metrics were Intersection over Mask Area and Intersection over Union, both in accuracy and in the average frame processing time. The analysis showed that the Intersection over Object Area and Convex Hull Intersection metrics solve the task most accurately. This determines the feasibility of using them for monitoring urban parking spaces.
Keywords:
OBJECT RECOGNITION, OBJECT INTERSECTION METRICS, CONVEX HULL INTERSECTION, INTERSECTION OVER UNION, PARKING SPACE, SMART PARKING
OBJECT RECOGNITION, OBJECT INTERSECTION METRICS, CONVEX HULL INTERSECTION, INTERSECTION OVER UNION, PARKING SPACE, SMART PARKING
