优化文件

This commit is contained in:
keita
2026-09-07 10:37:05 +08:00
parent b031b2e71a
commit e2dd4ab7f4
6 changed files with 7 additions and 216 deletions
+1 -1
View File
@@ -2,7 +2,7 @@
<module type="PYTHON_MODULE" version="4">
<component name="NewModuleRootManager">
<content url="file://$MODULE_DIR$" />
<orderEntry type="inheritedJdk" />
<orderEntry type="jdk" jdkName="D:\program file\conda\envs\pytorch" jdkType="Python SDK" />
<orderEntry type="sourceFolder" forTests="false" />
</component>
</module>
@@ -1,49 +0,0 @@
# C++ → Python 移植计划(reference-c++ → utils)
## 范围与环境
- 移植 `reference-c++/target_tracker` 与 `reference-c++/visual_track_control` 两个子项目为 Python,放入 `utils/`;**不改动 reference-c++ 任何文件**。
- Python 解释器:`D:\program file\conda\envs\pytorch\python.exe`(3.12.13,mavsdk 3.15.3 / numpy 2.4.3 已装)。需向该环境 `pip install pytest`。
- 行为保真优先:单位、符号约定、限幅顺序、超龄/判丢/Coast 语义、异常消息文本全部与 C++ 版及规格文档一致。
## 目录布局
```
utils/
├── README.md # 唯一文档(结构、运行、MAVSDK 映射决策、实机接线片段)
├── conftest.py # pytest sys.path 注入,使两包可顶层导入
├── target_tracker/ # 包1:追踪库(仅依赖 stdlib + numpy)
│ ├── __init__.py # 导出 TargetTracker/TrackResult/Detection/FrameShape 等
│ ├── kalman_box_filter.py # 8 维恒速卡尔曼(numpy 矩阵;Q 用状态值、R 用 abs×2 的规格差异保留)
│ ├── tracklet.py # 轨道 + ToCenterSize/ToXYXY/ComputeIoU
│ ├── target_tracker.py # 六步主流程 + 贪婪关联 + 主目标锁定;LastResult 用 threading.Lock
│ └── tests/ # pytest 移植 27 条(kalman 5 / tracklet 12 / tracker 10,含 §9.4 六条验收)
├── visual_track_control/ # 包2:控制律(不 import target_tracker)
│ ├── __init__.py
│ ├── vtc_types.py # TrackObservation/ControlSnapshot/ControlOutput/枚举(frozen dataclass)
│ ├── vtc_config.py # stdlib json 加载,校验异常消息与 C++ 一致(含字段路径)
│ ├── line_of_sight.py / pid_controller.py / rate_limiter.py / staleness_checker.py / frame_transform.py
│ ├── tracking_control_law.py # 八步单周期主流程,logging 仅模式切换打点
│ ├── offboard_sender.py # OffboardSender 协议 + YawIntegrator + LoopbackSender(演示/测试)
│ ├── mavsdk_sender.py # MAVSDK 实现(mavsdk 懒加载,其余模块零感知)
│ ├── config/track_control.json # 从 C++ 原样拷贝
│ └── tests/ # pytest 移植 8 套
└── examples/
└── track_control_demo.py # 唯一同时 import 两包的胶水点
```
## 关键设计
1. **解耦边界**:C++ 版 ControlSnapshot 内嵌 tracker::TrackResult;Python 改为控制律自有的最小结构 `TrackObservation(tracked, is_predicted, center)`。visual_track_control 全程不出现 target_tracker 导入,demo 负责字段拷贝。两包内部一律相对导入、可独立搬移;mavsdk 只在 mavsdk_sender.py 内懒加载。
2. **MAVSDK 映射(已确认方案)**:统一 `set_velocity_ned(VelocityNedYaw)`。
- 速率式:`YawIntegrator` 把 yaw_rate 积分为绝对航向(wrap180,dt 取发送侧单调时钟差)→ VelocityNedYaw(vn, ve, vd, 积分航向);
- 位置式:直接用 out.yaw_deg;刹车:零速 + 保持积分航向;降级:不发送。
- 控制律输出仍与 C++ 一致(机体系计算限幅→输出端旋转为 NED),NED 分量直接对应 MAVSDK 字段;OffboardError 捕获后返回 False(对齐 C++ Send 布尔语义)。offboard.start()/stop() 生命周期不属于 sender,README 给接线片段。
3. **数值**:全程 Python float(双精度),卡尔曼用 numpy(4×4 求逆 + 1e-9 正则化);保留检测入口 NaN/Inf 过滤、置信度 NaN→−∞ 排序等防御逻辑;时间戳 `time.monotonic_ns()//10**6`。
4. **MAVSDK sender 测试**:用 fake offboard 对象(鸭子类型)验证分发与积分行为,不 import mavsdk;gateway 相关 C++ 字节级用例改为验证 sender 收到的 VelocityNedYaw 内容。
## 验证与交付
1. `pip install pytest`(pytorch 环境)→ 两包测试全绿;
2. 运行 demo(asyncio,20Hz,三阶段:正常 6s → Coast 4s → 判丢 3s,演示判丢阈值 60 帧),输出语义与 C++ 版一致;
3. mavsdk_sender 冒烟测试(fake offboard,不连实机);
4. 撰写 utils/README.md。
## 实施顺序
pip 安装 pytest → target_tracker 包+测试 → visual_track_control 包+测试 → sender 层+测试 → demo → README → 汇总报告。
-163
View File
@@ -1,163 +0,0 @@
# code-test-drone
This project contains a C++ reference implementation of a visual target tracking and control system and its Python port, designed for drone tracking and control of moving targets.
## Project Structure
```
code-test-drone/
├── reference-c++/ # C++ reference implementation
│ ├── target_tracker/ # Target tracking library (independent sub-project)
│ │ ├── include/ # Header files directory
│ │ ├── src/ # Source files directory
│ │ └── tests/ # Unit tests
│ └── visual_track_control/# Visual tracking control library
│ ├── include/ # Header files directory
│ ├── src/ # Source files directory
│ └── tests/ # Unit tests
└── utils/ # Python port
├── target_tracker/ # Target tracking (Python version)
├── visual_track_control/# Tracking control (Python version)
└── tests/ # Unit tests
```
## Main Modules
### target_tracker - Target Tracking Library
Provides target tracking functionality based on Kalman filtering:
- **KalmanBoxFilter**: 8-dimensional Kalman filter (position, velocity)
- **Tracklet**: Target trajectory management and state updates
- **TargetTracker**: Main tracker, containing association and election logic
Core features:
- Greedy association algorithm for detection matching
- Three-level fallback mechanism for main target election
- Supports Coast (pure prediction) and Lost (missed) state management
### visual_track_control - Visual Tracking Control Library
Generates drone control commands based on tracking results:
- **PIDController**: Distance PID control
- **TrackingControlLaw**: Tracking control law
- **MavlinkCommandGateway**: MAVLink command gateway
- **PixelToAngle**: Pixel to angle conversion
- **StalenessChecker**: Data staleness check
- **SlewRateLimiter**: Rate limiter
Supports multiple control modes:
- **Rate Mode**: Angular rate control
- **Position Mode**: Absolute heading control
- **Distance Control**: Approach/retreat strategy
## Environment Requirements
### C++ Version
- CMake 3.10+
- C++17 Compiler
- MAVLink Library (MAVSDK)
### Python Version
- Python 3.12+
- Dependencies: `numpy`, `pymavlink`, `mavsdk`
## Installation
### Python Environment
```bash
pip install -r requirements.txt
```
### C++ Build
```bash
cd reference-c++/target_tracker
mkdir build && cd build
cmake ..
make
cd ../../visual_track_control
mkdir build && cd build
cmake ..
make
```
## Testing
### Python Tests
```bash
cd utils
pytest
```
### C++ Tests
```bash
cd reference-c++/target_tracker/build
ctest
cd ../../visual_track_control/build
ctest
```
## Configuration
Both C++ and Python versions use JSON format configuration files:
```json
{
"control": {
"frequency_hz": 20.0,
"visual_stale_ms": 200,
"radar_stale_ms": 500,
"attitude_stale_ms": 300
},
"camera": {
"image_width": 1920,
"image_height": 1080,
"fov_h_deg": 60.0,
"fov_v_deg": 35.0
},
"heading": {
"mode": "rate",
"gain": 1.0,
"yaw_rate_limit_dps": 30.0,
"yaw_slew_limit_dps2": 60.0
},
"distance": {
"d_exp_m": 10.0,
"kp": 0.5,
"ki": 0.05,
"kd": 0.1,
"integral_limit": 2.0
}
}
```
## Usage Examples
### Python Tracking Demo
```bash
cd utils
python demo_track_control.py
```
### C++ Tracking Demo
```bash
cd reference-c++/visual_track_control/build
./track_control_demo
```
## Porting Notes
Please refer to the README in each subdirectory for interface differences between the Python version and the C++ version:
- [target_tracker Python Version](./utils/target_tracker/README.md)
- [visual_track_control Python Version](./utils/visual_track_control/README.md)
## License
This project follows open source license agreements.
## Contributing Guidelines
Feel free to submit Issues and Pull Requests to improve this project.
-3
View File
@@ -1,7 +1,4 @@
抱歉,我无法访问代码仓库。让我直接根据提供的代码结构为你生成 README。
# code-test-drone
本项目包含一个视觉目标追踪与控制系统的 C++ 参考实现及其 Python 移植版本,用于无人机对移动目标的追踪与控制。
Binary file not shown.

After

Width:  |  Height:  |  Size: 113 KiB

+6
View File
@@ -0,0 +1,6 @@
## 1.已完成工作:
- 两C++模块移植到python
- mock测试通过
- 仿真测试一切正常
## 2.未完成工作
状态机代码尚未完成