UAV Optical Tracking System

A real-time, hardware-accelerated object detection and target tracking pipeline built in C++ for Unmanned Aerial Vehicle (UAV) camera systems.

Features

• AI Inference Pipeline: Executes a YOLOv8n ONNX model using OpenCV's DNN module, featuring custom output parsing strictly filtered to detect humans (Class 0) and cars (Class 2).
• Hardware Acceleration: Utilizes DNN_BACKEND_OPENCV and DNN_TARGET_OPENCL_FP16 for high-frame-rate performance natively optimized for OpenCL-compatible hardware, including the MacBook Air M3.
• Predictive Tracking: Implements a 4-state constant velocity Kalman Filter (position and velocity in 2D space) to predict trajectories and maintain target lock even through brief occlusions.
• Data Association & Gating: Associates raw detections with the Kalman state using a Nearest Neighbor algorithm based on Euclidean distance, applying a 100-pixel spatial gating threshold to reject false positives.
• Ground Station Telemetry: Includes a modular interface simulating a UDP communication channel to transmit live target coordinates and system lock states to a Ground Control Station.
• Advanced HUD: Renders real-time bounding boxes, crosshairs, Kalman-predicted trajectory markers, and system status overlays directly onto the video feed.

Architecture

• CMakeLists.txt: Build configuration supporting C++17 and OpenCV.
• src/main.cpp: Core application loop, dataset parsing, AI inference pipeline, and HUD rendering.
• src/UavTracker.hpp & src/UavTracker.cpp: Kalman filter logic and nearest neighbor target association implementation.
• src/Telemetry.hpp: Distributed communication architecture mockups for UAV-to-GCS links.

Prerequisites

• Compiler: C++17 compatible compiler.
• Build System: CMake 3.16+.
• Dependencies: OpenCV compiled with OpenCL and DNN module support.

Build Instructions

1. Clone the repository and initialize the build directory:

   mkdir build
   cd build
   cmake ..
   make

   This compiles the source and generates the tracker executable.

Usage & Configuration

Prior to executing the tracking pipeline, update the absolute paths in src/main.cpp to target your local data:

1. Image Sequence: Specify the path and zero-padding format for the sequential dataset (e.g., Kaggle/VisDrone sequences):

   std::string kaggleSequencePath = "/Users/alidai/Downloads/KaggleDataset/images/frame_%06d.jpg";

2. Model File: Set the correct path to your downloaded yolov8n.onnx file:

   cv::dnn::Net net = cv::dnn::readNetFromONNX("/Users/alidai/Desktop/opencv-gpu-tracker/models/yolov8n.onnx");

Once configured and built, run the executable:

./tracker