Amidst the diverse applications, such as airport security, shopping mall monitoring, and public space surveillance, this project delves into the challenges of detecting pedestrians and tracking their movements in surveillance footage.
The "Detection and Tracking of Pedestrians" project, developed as part of the Visual Computing course (CVI) in the academic year 2020/2021, presents a system for detecting pedestrians and tracing their movements within surveillance video footage. This endeavor showcases the synergy of computer vision techniques and video analysis, contributing to advancements in surveillance and crowd behavior analysis.
The project is centered around designing and implementing a video surveillance system that can accurately detect pedestrians and track their trajectories across consecutive frames. By harnessing the power of computer vision, the system identifies key points of interest within the video and traces the movement patterns of pedestrians over time. This application has implications for security, crowd management, and public safety.
The project utilizes the following stack:
- Matlab and Python for implementing computer vision algorithms and analysis.
| Name | ISTID |
|---|---|
| Catarina Rodrigues | 87817 |
| Daniela Mendes | 87646 |
| Vasco Pires | 87708 |
To validate the algorithm's effectiveness, the project draws upon benchmark datasets, particularly the PETS family dataset. The chosen subset, S2.L1, provides diverse sequences and viewpoints of pedestrians in varying environments. Each sequence is accompanied by ground truth data, encompassing object locations and sizes (bounding boxes) for each frame.
The project dives into several critical tasks, including:
- Implementation of pedestrian tracking with visible bounding boxes for detections.
- Dynamic plotting of pedestrian trajectories for clear visualization.
- Integration of pedestrian labeling to facilitate identification.
- Provision of a map illustrating pedestrian trajectories within the video.
- Exploration of heatmap generation based on the frequency of occurrences in distinct image regions.
- Investigation into the incorporation of optical flow information to capture pedestrian motion.
To run the project follow these steps:
- MATLAB - Get the Software
- Run
start_here.mfrom the project directory.- A numbered menu will appear, presenting various options for pedestrian detection and tracking.
- Select an appropriate option to inspect pedestrians within the surveillance footage.
- Visualize detected pedestrians and their trajectories, along with additional functionalities for video analysis.
This project showcases the integration of visual computing concepts to solve real-world challenges. By successfully developing a system that can detect and track pedestrians, the project contributes to the field of surveillance technology. The ability to analyze crowd behavior and pedestrian movement patterns has far-reaching implications in areas such as security, urban planning, and public safety. Thank you for exploring our project!