Skip to content

Latest commit

 

History

History
42 lines (30 loc) · 2.72 KB

README.md

File metadata and controls

42 lines (30 loc) · 2.72 KB

Yolov5-Human-Tracking

This project utilizes the YOLOv5 object detection algorithm and the DeepSORT object tracking algorithm to detect and track people in a video. By analyzing the movement trajectories of each person in the video, the project can calculate their dwell time and other related information.

Method Overview:

  1. Perform frame-by-frame analysis of the video using the YOLOv5 algorithm to recognize and detect people in the video.

  2. Apply the DeepSORT algorithm to track the detected people, assigning a unique ID to each person.

  3. Calculate the center point of each person and add it to the tracking trajectory. Trajectories are stored as dictionaries, with the person's ID as the key.

  4. Draw bounding boxes and trajectory lines for each person in the video.

  5. Periodically save information such as the trajectory length, dwell time, and average speed of the people to a CSV file (e.g., every 10 seconds).

  6. Facial Blurring using MHCNN (Multi-Task Hierarchical Convolutional Neural Network) to blur faces for privacy protection. However, if you are using an infrared camera, this step can be skipped.

  7. Save images containing the person's trajectory and images containing only the trajectory as image files.

1-01 2-01

Required Libraries:

  • os: For operating system-related functions.
  • time: For handling time-related functions.
  • numpy: For numerical operations and array manipulation.
  • detect: Custom module for object detection using YOLOv5.
  • cv2 (OpenCV): For image and video processing.
  • csv: For reading and writing CSV files.
  • deep_sort: Custom module for object tracking using the DeepSORT algorithm.
  • collections: For using deque data structure.
  • yaml: For parsing YAML configuration files.

Result Demonstration:

4_画板 1

Camera usage advice:

  1. 720p Webcam (RGB): Ample outdoor lighting, allowing for capturing full-body shots.
  2. 120p Infrared Camera (Gray): An indoor environment with low room temperature, or an outdoor environment at night. Avoid brightly lit situations.
  3. 120p Infrared Camera (Normal): An indoor environment with low or medium room temperature, or an outdoor environment at night. Avoid brightly lit situations

3-01

PS:Thanks to George Verghese and Jam Kim for their contributions during the video collection