Price: 0
Number of applications: 14
26.05.26 (inclusive)
Monetary
MVP
ICT tasks
Food industry
Neurotechnology and artificial Intelligence
Software/ IS
Currently, monitoring the presence of employees and detecting suspicious activity in the shopping area requires manual monitoring of video cameras or subsequent viewing of recordings. This approach takes a lot of time, depends on the human factor, and does not allow for prompt response to prolonged absence of an employee from the workplace or potentially suspicious behavior of visitors. Existing surveillance cameras record what is happening, but they do not analyze events by themselves: they do not identify people in the frame, do not track their movement between zones, do not record the appearance or absence of an employee, and do not generate automatic notifications on potentially important situations. An additional complication lies in the fact that video quality, distance to people, camera viewing angles, and stream stability directly affect the possibility of correct AI processing. If the image is not clear enough or the person is poorly distinguishable in the frame, the system may lose the object, confuse people between cameras, or generate inaccurate events. Also, the identification of suspicious actions related to possible thefts cannot be completely accurate at the first stage, since it requires the accumulation of real data, confirmation of events by people and subsequent fine-tuning of logic. Therefore, the problem lies not only in the automation of video surveillance, but also in the need to create an MVP system that can start collecting data, record basic events and gradually improve the accuracy of the analysis.
Automation of video monitoring, reduction of dependence on manual camera viewing, basic monitoring of the presence of employees and the formation of initial notifications of suspicious activity.
Natalia Alexandrovna Chuprun
Purpose and description of task (project)
The purpose of the project: To develop and implement an MVP AI video monitoring system based on existing video surveillance cameras for basic monitoring of the presence of employees, recording their absence, approximate arrival/departure time, as well as the initial detection of suspicious behavior in the shopping area. Task/Project description: As part of the project, it is planned to connect video streams from cameras via RTSP, check image quality and configure the basic video processing pipeline. The system should identify people in the frame, perform basic tracking of the same person within the surveillance area, and work with multiple cameras without mixing objects between streams. At the first stage, the system will be used to monitor employees without accurate identification: by appearance, uniform or clothing. The system will record the fact of a person's appearance in the work area, prolonged absence, as well as the approximate time of arrival and departure for the first and last appearance in the frame. At the next stage, the basic logic for detecting suspicious behavior will be implemented at the level of simple heuristics: moving a hand towards a body or bag, staying at a shelf with your back to the camera for a long time, and checking the event based on data from another camera if it sees the same scene from a different angle. Additionally, it provides for sending notifications to the Telegram group, collecting feedback on the confirmation or rejection of events, as well as using this data to further improve the logic of the system. At the MVP stage, suspicious activity detection will have limited accuracy and will require subsequent fine-tuning based on real data. The quality of the system's operation directly depends on the quality of the cameras, the distance to objects, viewing angles, and the stability of video streams.
Whatsapp
