Ai People Tracking | AI Final Year Project Report

Complete Technical Paper Content — Ai People Tracking

1. Executive Abstract, Problem Formulation & Background

MVIG-SJTU / AlphaPose Star 8.6k Real-Time and Accurate Full-Body Multi-Person Pose Estimation&Tracking System tracking skeleton gpu realtime pytorch keypoints human-pose-estimation human-computer-interaction pose-estimation accurate human-tracking posetracking alphapose human-pose-tracking alpha-pose person-pose-estimation crowdpose full-body whole-body human-joints Updated May 13, 2024 Python

m-dorgham / PedestrianTracking Star 98 People detection and tracking in stationary RGB cameras people-detection pedestrian-tracking human-tracking Updated Sep 30, 2017 C++

lewjiayi / Crowd-Analysis Star 94 System integrated with YOLOv4 and Deep SORT for real-time crowd monitoring, then perform crowd analysis. The system is able to monitor for abnormal crowd activity, social distance violation and restricted entry. The other part of the system can then process crowd movement data into optical flow, heatmap and energy graph. python yolo movement-detection crowd-analysis human-tracking movement-data-analysis deep-sort social-distancing yolov4 abnormal-crowd-activity crowd-movement crowd-monitoring Updated Dec 15, 2021 Python

System integrated with YOLOv4 and Deep SORT for real-time crowd monitoring, then perform crowd analysis. The system is able to monitor for abnormal crowd activity, social distance violation and restricted entry. The other part of the system can then process crowd movement data into optical flow, heatmap and energy graph.

dasupradyumna / YOLOv7-DeepSORT-Human-Tracking Star 39 A human detector and tracker, written in python, using YOLOv7 for detection and DeepSORT for tracking the detections from YOLO. computer-vision deep-learning pytorch neural-networks yolo machine-vision human-tracking deepsort yolov7-deepsort Updated Feb 12, 2023 Python

A human detector and tracker, written in python, using YOLOv7 for detection and DeepSORT for tracking the detections from YOLO.

LCAS / FLOBOT Star 27 EU funded Horizon 2020 project robotics ros dataset perceptron velodyne rgbd h2020 environmental-modelling human-detection 3d-lidar human-tracking Updated Nov 8, 2022 C++

anand-b / Real-time-Person-Tracking-and-Height-Estimation Star 18 The project tracks a person wearing a specific logo t-shirt from video and estimates height of the person computer-vision video-processing opencv-python human-tracking Updated May 28, 2018 Python

2. Related Work, Literature Survey & Algorithmic Foundations

The project tracks a person wearing a specific logo t-shirt from video and estimates height of the person

ozora-ogino / tflite-human-tracking Star 12 SORT based human tracking for edge. yolo object-detection object-tracking human-tracking tflite deepsort yolov5 Updated Mar 12, 2022 Python

AdrianPrados / TAICHI Star 11 This is the repository for the TAICHI algortithm for non-antropomorphic robotic arms. learning matlab manipulation lfd imitation-learning mujoco human-tracking ur3 Updated Nov 13, 2023 Python

RuizeHan / CVMHT Star 10 CVMHT : Complementary-View Multiple Human Tracking (AAAI 2020). multi-object-tracking human-tracking multi-view Updated Dec 9, 2021 MATLAB

muki01 / OpenCV_Projects Star 10 OpenCV projects like Face Recognition, Item Detection, Hand Tracking, Body Detection, Color Detection and more. opencv face-recognition face-detection opencv-python hand-tracking color-detection human-detection hand-detection human-tracking item-detection Updated Apr 9, 2024 Python

OpenCV projects like Face Recognition, Item Detection, Hand Tracking, Body Detection, Color Detection and more.

PanithanS / Webcam-Pose-Estimation-Walking-Speed-Measurement Star 10 A prototype with a demonstration video for real-time human pose estimation and walking speed measurement using YOLOv8 with webcam. opencv computer-vision object-detection human-pose-estimation walking pose-estimation human-tracking speed-monitor pose-tracking pose-detection yolov8 walking-speed Updated Mar 28, 2024 Python

A prototype with a demonstration video for real-time human pose estimation and walking speed measurement using YOLOv8 with webcam.

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

amajji / real-time-human-detection-tracking-system Star 9 Real-time human detection and tracking camera using YOLOV5 and Arduino arduino camera websockets arduino-ide unittest webapp yolo arduino-uno workflows human-detection human-tracking human-detection-algorithm fastapi github-workflows yolov5 smart-camera Updated Nov 26, 2023 Python

hamzashafiq28 / Object-Detection-and-Tracking Star 8 Human Tracking System: Multi-tracker detection and tracking system with YOLOv12 and BoxMOT, optimized for dance videos with mobile deployment capabilities. computer-vision pytorch object-detection collaborate object-tracking multiple-object-tracking human-tracking yolov12-object-tracking boxmot Updated Apr 15, 2025 Python

Human Tracking System: Multi-tracker detection and tracking system with YOLOv12 and BoxMOT, optimized for dance videos with mobile deployment capabilities.

eduze / AugurSense Star 7 Camera based Ground Mapped Real-time Analytics on Movements of People tracking analytics image-processing vision pose-estimation re-identification human-tracking Updated Dec 30, 2022 HTML

Yazdi9 / 4D_HumanBodyReconstructing Star 6 4D& 3D Reconstructing and Tracking Humans Bodies In videos computer-vision video-processing mesh-generation 3d-reconstruction human-tracking 4d-reconstruction human-reconstruction Updated Oct 4, 2024 Jupyter Notebook

ricnah / SWEENEX Star 5 📡 WiFi-based 3D human pose estimation using Channel State Information (CSI) — no camera, no wearable. ESP32 + existing home routers + Python + Next.js. Real-time skeleton tracking in your browser via WebGL. Privacy-first, sub-$5 hardware. open-source iot machine-learning privacy real-time embedded deep-learning websocket nextjs esp32 transformer csi pose-estimation channel-state-information human-tracking onnx wifi-scanner 3d-skeleton indoor-localization wifi-sensing Updated May 11, 2026 TypeScript

📡 WiFi-based 3D human pose estimation using Channel State Information (CSI) — no camera, no wearable. ESP32 + existing home routers + Python + Next.js. Real-time skeleton tracking in your browser via WebGL. Privacy-first, sub-$5 hardware.

saigiricharan / humanoid-edge-ai-human-tracking-robot Star 3 Humanoid-inspired autonomous robot using Edge AI, ROS2, Raspberry Pi, ESP32, and YOLOv8 for real-time human detection and tracking. opencv computer-vision robotics raspberrypi esp32 embedded-systems ros2 human-tracking onnx edge-ai autonomous-robot yolov8 Updated Jun 18, 2026 Python

4. Neural Network Backbone, Layer Specifications & Loss Formulations

Humanoid-inspired autonomous robot using Edge AI, ROS2, Raspberry Pi, ESP32, and YOLOv8 for real-time human detection and tracking.

kajuyerim / YOLOv8-and-DEEPSORT-Human-Foot-Trafficker Star 3 Computer vision project for human foot trafficking with data categorization. human-tracking deepsort yolov8 Updated Apr 29, 2024 Python

Akshit0203 / wifi-csi-human-radar Sponsor Star 2 WiFi CSI Human Presence Detection and Tracking System using ESP32-S3, Python, WebSockets, and Real-Time 3D Visualization. python iot threejs real-time signal-processing esp32 websockets activity-recognition embedded-systems presence-detection human-detection human-tracking esp32-s3 wireless-sensing Updated Aug 2, 2026 HTML

WiFi CSI Human Presence Detection and Tracking System using ESP32-S3, Python, WebSockets, and Real-Time 3D Visualization.

Kaminyou / Movement-Anywhere Star 2 [IEEE JBHI 2026] Official implementation of "Movement Anywhere: An Open-Source Distributed 2D Video-Based Movement Analysis Platform Empowered by Active Learning" distributed-systems distributed-computing active-learning pose-estimation depth-estimation gait-analysis human-tracking movement-analysis Updated May 12, 2026 Python

[IEEE JBHI 2026] Official implementation of "Movement Anywhere: An Open-Source Distributed 2D Video-Based Movement Analysis Platform Empowered by Active Learning"

To associate your repository with the human-tracking topic, visit your repo's landing page and select "manage topics."

Tracking People using YOLOv7 and DeepSORT Table of Contents Overview Getting Started Code External Work Overview This repository implements a solution to the problem of tracking moving people in a low-quality video. It uses a state-of-the-art object detector YOLOv7 for detecting people in a frame, and fuses these robust detections with the bounding boxes of previously tracked people using the neural network version of SORT called DeepSORT tracker. The packages for YOLO and DeepSORT algorithms are located under yolo and deepsort folders, where the modules are adapted from the official repositories to fit the scripts here. Getting Started The necessary python environment required for working with this repository can be setup using the script setup_conda.sh . This script assumes that the system has a working conda distribution (Miniconda or Anaconda), and accepts one argument which will be used as the name for the newly created environment. The installed conda environment will have the following major packages: TensorFlow 2.8.1 (with GPU support) CUDA 10.2 (with CuDNN 7.6.5) PyTorch 1.10.1 (with TorchVision 0.11.2) OpenCV 4.6.0 Matplotib 3.6.3 Pandas 1.5.3 Note (1) : The script is not robust, and may fail when conda packages are updated in their respective channels. Note (2) : In case of import errors, it is recommended to have "channel_priority" set to "true" in ".condarc" file, with the channels priority being "pytorch" > "conda-forge" > "defaults", and rerun the setup script. Code After setting up the conda environment using the above script, the pipeline can be executed by - python main.py --input-vid=/path/to/input/video/file --save-path=/path/to/output/video/file This will read frames from the input video, detect people using a YOLO API, and track people using a DeepSORT API. These APIs can be found in api folder. The frames with people detected and tracked are saved to the specified video file. Data and Results Input video, versioned output videos, checkpoints and a README with versioning details can be found here . The checkpoints folder should be downloaded to the project root. External Work YOLOv7 Repository - Paper All the modules related to this network can be found under yolo/ , and the weights can be found at checkpoints/yolov7x.pt in the GDrive link. More details can be found in yolo/__init__.py . DeepSORT Repository - Paper All the modules related to this network can be found under deepsort/ , and the weights can be found at checkpoints/ReID.pb in the GDrive link. More details can be found in deepsort/__init__.py .

5. Training Protocol, Convergence Behavior & Experimental Evaluation

This repository implements a solution to the problem of tracking moving people in a low-quality video. It uses a state-of-the-art object detector YOLOv7 for detecting people in a frame, and fuses these robust detections with the bounding boxes of previously tracked people using the neural network version of SORT called DeepSORT tracker. The packages for YOLO and DeepSORT algorithms are located under yolo and deepsort folders, where the modules are adapted from the official repositories to fit the scripts here.

The necessary python environment required for working with this repository can be setup using the script setup_conda.sh . This script assumes that the system has a working conda distribution (Miniconda or Anaconda), and accepts one argument which will be used as the name for the newly created environment. The installed conda environment will have the following major packages:

Note (1) : The script is not robust, and may fail when conda packages are updated in their respective channels. Note (2) : In case of import errors, it is recommended to have "channel_priority" set to "true" in ".condarc" file, with the channels priority being "pytorch" > "conda-forge" > "defaults", and rerun the setup script.

After setting up the conda environment using the above script, the pipeline can be executed by -

python main.py --input-vid=/path/to/input/video/file --save-path=/path/to/output/video/file

This will read frames from the input video, detect people using a YOLO API, and track people using a DeepSORT API. These APIs can be found in api folder. The frames with people detected and tracked are saved to the specified video file.

Input video, versioned output videos, checkpoints and a README with versioning details can be found here . The checkpoints folder should be downloaded to the project root.

All the modules related to this network can be found under yolo/ , and the weights can be found at checkpoints/yolov7x.pt in the GDrive link. More details can be found in yolo/__init__.py .