A prototype that detects people with YOLOv8 and tracks them with DeepSORT in defined work areas, counts presence time in seconds, and exports records to Excel.
An academic project carried out by a team of students with technical assistance from Techno Enjaz to develop a prototype that uses computer vision to detect and track people inside defined work areas, calculate how long each tracked person remains within the area, and then organize the temporal data and export it as reviewable reports.
For implementation, the prototype relied on YOLOv8 for person detection and DeepSORT for tracking people across video frames, together with software logic that computes presence duration and links the results to structured data that can be exported to an Excel file. The experiment was run on a video clip simulating an office environment, so the results represent a functional proof of concept for a prototype rather than a production deployment inside an actual company.
| Item | Details |
|---|---|
| Project type | Computer vision and video analytics prototype |
| Field | Artificial intelligence, computer vision, person tracking |
| Project status | Prototype tested on a video simulating an office environment |
| Techno Enjaz role | Technical support and assistance to the students during project development |
| Core technologies | YOLOv8, DeepSORT, OpenCV, pandas, datetime |
| Outputs | Person tracking, defined work areas, presence-time calculation, a database (ID, office number, duration, date), export to Excel |
Many companies rely on fingerprint readers or smart cards to record arrival and departure. These methods capture only the moments of entering and leaving; they do not show whether an employee is actually present in their work area during the shift, or has left it for long periods without permission. The project starts from this gap: instead of recording two timestamps, it tries to measure actual time spent inside the work area throughout the recording, using a camera and video analysis.
The project focused on building a model capable of analyzing video of a work environment, detecting the people in the scene, and then tracking each person's movement across frames using an independent tracking ID. The tracked positions are then compared against predefined work areas, and the time each tracked person spends inside their area is computed.
The goal of the prototype is to test whether computer vision techniques can automate part of the attendance-recording process inside work areas, while providing temporal data that can be reviewed administratively. The current version does not rely on face recognition to confirm employee identity; it focuses on tracking people in the scene using tracking IDs, while linking those IDs to a confirmed employee identity remains an area for subsequent development.
The project was carried out academically by the students, with technical assistance and support from Techno Enjaz during the development phases. It is presented among the office's work as an example of its contribution to supporting applied projects that combine artificial intelligence, computer vision, and video analytics.
This page does not attribute full project execution or independent training of all models to Techno Enjaz; the office's role in this project was technical assistance to the students during development.
According to the project documentation, implementation went through the following stages:
The prototype goes through several interconnected stages:
The project presents screenshots of the system running on the office-scene video, showing the following elements:
| Visual element | Meaning |
|---|---|
| Red boxes around people | People detected by the model in the scene |
| Green work-area borders | Predefined work areas, each with an office number |
| Seconds counter | Time each person has spent in their area |
| Area border turning red | The associated person has left the work area, as in the example of the employee who left area 5 |
The results are saved in a table containing the employee ID, office number, working duration, and date, and then exported to an Excel file.
Among the challenges the project addressed are people overlapping in the scene and temporary loss of tracking — well-known problems in multi-object tracking systems. Moving from an experimental model to a real application also requires handling additional variables, such as lighting variation, crowd density, multiple cameras, and maintaining a tracked person's identity when they leave the scene and return.
Another important aspect is distinguishing between a tracking ID and an actual employee identity. The documented version of the project tracks people within the video, but it does not provide evidence of an independent identity-verification layer such as face recognition or employee ID cards.
The experiment demonstrated that the following core functions can be implemented within the test scenario:
These results represent functional success of the prototype and are not a measurement of a production system's accuracy. The source did not report project-specific quantitative values such as Precision, Recall, mAP, FPS, IDF1, or ID-switch rate, so no undocumented accuracy or speed figures are attributed to the current version.
Used to detect people within the video frames and determine their positions in the scene.
Used to track people across frames and assign tracking IDs that allow each person's movement to be followed for as long as they appear in the video.
Used within video and frame processing and for displaying the analysis results visually.
Used to organize the data and handle the temporal values needed to compute presence duration and prepare the records.
The project was designed as an academic prototype, so its results should be read within that frame:
The project offers hands-on experience in integrating person detection, multi-object tracking, regions of interest, time computation, and data management within a single pipeline. Its value as an applied study lies in moving from detecting a person in an image to following their presence over time and tying that presence to a specific area and a reviewable record.
For Techno Enjaz, the project reflects part of the office's experience in supporting applied projects built on computer vision and artificial intelligence, while maintaining a clear distinction between the office's contribution and the academic role of the students who carried out the project.
According to the development directions stated in the project, the prototype could be extended in the future through:
These items remain future developments and are not functions attributed to the current version.
Because this type of system processes video of people inside a work environment, any actual use requires clear controls covering consent, the purpose of the monitoring, access permissions, data retention periods, and the protection of images and records.
If your project needs video analytics, person or object detection, tracking, or a computer-vision-based prototype, you can contact Techno Enjaz to discuss the technical requirements and the appropriate scope of support for the project.
A prototype built by a team of students with technical assistance from Techno Enjaz that analyzes workplace video to detect people, track them inside defined work areas, and compute how long each tracked person stays in their area. Results are stored with the office number and date and exported to an Excel file for review.
YOLOv8 detects people in each frame, DeepSORT keeps a tracking ID for each person across frames, and each position is compared with the predefined work area. Timing logic built on datetime and pandas computes presence in seconds, and the area border turns red when its occupant leaves.
Not in the current version. It relies on tracking IDs within the video, and a tracking ID does not automatically equal a confirmed employee identity. The project lists face recognition as a future development, not a proven feature.
The prototype was tested on a video simulating an office scene and worked functionally: detecting and tracking people, computing their presence time in their areas, flagging when one left their area, and exporting data to Excel. The source reports no quantitative metrics such as Precision, mAP, FPS, or IDF1, so no accuracy figures are attributed to it.
It would need testing under real operating conditions with varied lighting and crowd density, a reliable employee identity-verification layer, and measurement of detection, tracking, and timing accuracy. It also requires clear privacy controls covering consent, the purpose of monitoring, access rights, and retention of recordings.
Reach out to Techno Enjaz engineering team to discuss technical feasibility, prototype hardware, and development scope.
Discover more engineering prototypes and academic implementations
A virtual board for writing, drawing, picking colors and erasing with hand gestures in front of a camera, built in Python with MediaPipe hand tracking and OpenCV.
An exam-monitoring prototype using YOLO, MediaPipe and OpenCV to analyze a student's head direction, flagging head turns in red for review and forward gaze in green.
A face-recognition access control prototype using OpenCV and face-recognition that logs authorized entries, saves images of unknown faces, and sounds an alarm.
