Person Detection(Yolov8x) & Tracking(DeepSORT) in Video Streams.

Overview

This project aims to detect and count the number of distinct person in a given input video stream using Yolov8x for object detection and DeepSORT object tracking.

Execution Configuration

• GPU: NVIDIA RTX 3070
• Model: YOLOv8x for object detection + DeepSORT for object Tracking
• Python Version:: 3.11.10

For detailed list of packages used, check requirements.txt

Program Descriptions + Sample Outputs

1. Simple object detection using Yolov8x.pt model (Source: yolov8x.py)

2. Detection + Tracking using Interaction Over Union (IOU) -- Highly inaccurate tracking mechanism. (Source: yolov8x_plus_iou.py)

3. Detection + Tracking using SORT Tracker https://github.com/abewley/sort -- Significantly more accurate than IOU. (Source: yolov8x_plus_sort.py)

4. Detection + Tracking using DeepSORT Tracker https://github.com/nwojke/deep_sort -- Again, a huge improvement over SORT. (Source: yolov8x_plus_deepsort.py)

5. Additional Footage:

GPU Load Measured while Inferencing (Notice 1200MHz and VRAM gets pushed to 7800MHz)

Data Sources

• https://motchallenge.net/data/MOT16/
• https://www.youtube.com/watch?v=WvhYuDvH17I
• https://www.youtube.com/watch?v=ZvzKuqSDyG8