GRMCF: Generalizable-Rapid-Multi-Target-Capture-Framework-Based-on-a-Single-Conventional-PTZ-Camera

A generalizable rapid multi-target capture framework based on a single conventional PTZ camera.

This repository provides the official open-source code for the paper with DOI: 10.1016/j.eswa.2026.131189.

Workflow figure (PDF): ./workflow_figure.pdf

---

Key Components

1) Feasible Viewing Volume Back-Projection Modeling

Given wide-angle target bounding boxes, GRMCF back-projects them into the PTZ parameter space and constructs feasible “viewing volumes” (geometric neighborhoods) for each target, enabling constraint-aware scheduling.

2) Fast Heuristic Optimization Algorithm (OGSOA-GRO)

GRMCF uses OGSOA-GRO to efficiently optimize:

• the visiting order (sequence),
• and the concrete PTZ states (pan, tilt, zoom) under 3D geometric constraints, to produce a rapid capture schedule suitable for real-time settings.

---

Repository Layout

• core/ Includes the complete source code and a runnable demo for the two core GRMCF components:

  • Modeling program: constructs feasible viewing volumes from target detections and PTZ camera parameters.
  • OGSOA-GRO solver: optimizes the PTZ capture sequence and executable PTZ states.

  The modeling program and the OGSOA-GRO solver are implemented as two independent programs and are both invoked through HTTP APIs.

• ablation_studies/
  Includes:

  • GRMCF core Linux executables (modeling + OGSOA-GRO solver)
  • All ablation experiment code
  • Synthetic data generation code

• ptz_time_cost_calibration/
  Includes:

  • PTZ time-cost calibration programs (pan/tilt/zoom/autofocus stabilization fitting)
  • Real measured calibration data from Hikvision DS-2DC4A423IW-DE(S6)
  • Scripts/functions to reproduce fitted time-cost curves

• workflow_figure.pdf
  A workflow PDF that presents the GRMCF pipeline (from detections to an executable PTZ schedule).

---

Workflow Overview

A typical capture workflow is:

1. Input: target detections as 2D bounding boxes in the wide-angle image
2. Back-projection modeling: map each target box to feasible PTZ neighborhoods (viewing volumes)
3. Heuristic scheduling (OGSOA-GRO): jointly optimize sequence + feasible PTZ states
4. Output: a PTZ control sequence [(P1, T1, Z1), (P2, T2, Z2), ...] for rapid capture

---

Getting Started

Clone

git clone https://github.com/jiaofangxu/GRMCF.git
cd GRMCF

For the full PTZ simulation source code, including the UE4 project files (>10 GB), please contact the paper author at xueshengjiaofangxu@163.com.