Monitoring the condition of electric scooter parking lots
About project
The task was set to create a system that allows you to record violations when parking electric scooters: disorder, overflow, parking in the wrong place, abandoned scooters, fire.
Realization
2.
Initial experiments and preparation of additional data
1.
Анализ проблемы и данных
3.
Preparation of an additional model and postprocessing logic
Testing. Modification.
5.
4.
Final training of models. Integration.
Conclusions based on the work done
The problem of determining violations related to electric scooters and their parking is solved by computer vision methods.
Data plays a crucial role in the quality of the system. Full-fledged training of models requires at least 100 snapshots from at least 100 cameras taken at different times of the day, in different weather and at different times of the year. Cameras must have a resolution of at least 1280*720, while the vertical size of the detected objects must be at least 5% of the vertical resolution of the camera, optimally at least 10%. Also, for the system to work, it is necessary to mark official parking areas with rectangular boxes in the coordinates of camera snapshots.
Based on the available data, an accuracy of 73% was achieved with a completeness of 72%. Identification of deviations from plans
During testing, single false detections are noted. It requires parameter setting and additional training on a large array of data.
After the system is put into operation, data collection will be required (provided in the system) for further marking and further training of models.