在 RDK X5 的 10TOPS BPU 上运行单个 AI 推理算法:YOLO 目标检测、图像分类、语义分割、人脸识别、手势识别、人体关键点、开放词汇检测(DOSOD/YOLO-World)、双目深度估计、语音识别、端侧轻量 LLM(≤2B 参数量化模型)。Use when the user wants to run a single AI algorithm, deploy pre-built .bin models to BPU, or ask what AI models/LLMs RDK X5 can run (能力咨询). Do NOT use for camera hardware setup (use rdk-x5-camera), multimedia encoding (use rdk-x5-media), running /app demo scripts (use rdk-x5-app), integrated camera+AI+output pipeline (use rdk-x5-tros), model conversion/quantizatio...
Scanned 9/9/2026
Install to Claude Code
npx -y skills add Lord1Egypt/awesome-skill-forge --skill rdk-x5-ai-detect --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Rdk X5 Ai Detect?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/lord1egypt-rdk-x5-ai-detect)More formats (shields.io, HTML) on the badges page.
---
name: rdk-x5-ai-detect
description: "在 RDK X5 的 10TOPS BPU 上运行单个 AI 推理算法:YOLO 目标检测、图像分类、语义分割、人脸识别、手势识别、人体关键点、开放词汇检测(DOSOD/YOLO-World)、双目深度估计、语音识别、端侧轻量 LLM(≤2B 参数量化模型)。Use when the user wants to run a single AI algorithm, deploy pre-built .bin models to BPU, or ask what AI models/LLMs RDK X5 can run (能力咨询). Do NOT use for camera hardware setup (use rdk-x5-camera), multimedia encoding (use rdk-x5-media), running /app demo scripts (use rdk-x5-app), integrated camera+AI+output pipeline (use rdk-x5-tros), model conversion/quantization/export (use Horizon toolchain on PC), large generative models like Stable Diffusion (insufficient VRAM), or custom code that bridges AI results to external services like MQTT."
license: MIT-0
metadata:
openclaw:
requires:
bins:
- ros2
compatibility:
platform: rdk-x5
---
# RDK X5 AI Detect — BPU AI 推理
RDK X5 搭载 10 TOPS BPU,所有 AI 算法通过 TogetheROS.Bot (tros.b) 框架运行。
## 前置准备
```bash
# 每个终端都需要 source
source /opt/tros/humble/setup.bash
# 安装 Python 推理库(可选,用于自定义推理脚本)
pip3 install hobot-dnn-rdkx5 hobot-vio-rdkx5
```
## 操作步骤
### 1. YOLO 目标检测(摄像头实时)
```bash
source /opt/tros/humble/setup.bash
ros2 launch dnn_node_example dnn_node_example.launch.py \
dnn_example_config_file:=config/yolov5sworkconfig.json \
dnn_example_image_width:=640 dnn_example_image_height:=480
```
浏览器 `http://<RDK_IP>:8000` 查看检测结果。
可选模型配置文件:
| 模型 | config 文件 |
|------|------------|
| YOLOv5s | `config/yolov5sworkconfig.json` |
| MobileNetV2 分类 | `config/mobilenetv2workconfig.json` |
### 2. 图像分类
```bash
source /opt/tros/humble/setup.bash
ros2 launch dnn_node_example dnn_node_example.launch.py \
dnn_example_config_file:=config/mobilenetv2workconfig.json
```
### 3. 开放词汇检测(DOSOD / YOLO-World)
```bash
source /opt/tros/humble/setup.bash
# DOSOD(地瓜自研,支持语音指令指定检测目标)
ros2 launch hobot_dosod dosod.launch.py
# YOLO-World
ros2 launch hobot_yolo_world hobot_yolo_world.launch.py
```
### 4. 双目深度估计
```bash
source /opt/tros/humble/setup.bash
# 需配合双目摄像头(SC230ai / SC132gs / ZED)
ros2 launch hobot_stereonet stereonet_model.launch.py
```
### 5. 语音识别 (ASR)
```bash
source /opt/tros/humble/setup.bash
ros2 launch hobot_asr hobot_asr.launch.py
```
需配合音频子板(微雪 WM8960 Audio HAT 或幻尔载板)。
### 6. 端侧 LLM 大模型
```bash
source /opt/tros/humble/setup.bash
ros2 launch hobot_llamacpp hobot_llamacpp.launch.py
```
### 7. Web 可视化
```bash
source /opt/tros/humble/setup.bash
ros2 launch websocket websocket.launch.py
```
所有算法结果叠加实时画面,浏览器 `http://<RDK_IP>:8000` 查看。
### 8. RTSP 拉流 + AI 推理(智能盒子方案)
```bash
source /opt/tros/humble/setup.bash
ros2 launch hobot_rtsp_client hobot_rtsp_client.launch.py \
rtsp_url:="rtsp://admin:password@192.168.1.64:554/stream1"
```
## 性能监控
```bash
# BPU 使用率
cat /sys/devices/system/bpu/bpu0/ratio
# 推理帧率
ros2 topic hz /ai_msg
```
## 排查故障
| 现象 | 原因 | 解决 |
|------|------|------|
| `Load model failed` | 模型文件不存在或格式错误 | 检查 .bin 文件路径;模型需经 hb_mapper 转换为 NV12 格式 |
| BPU ratio 始终 0 | 推理节点未启动或模型未加载 | `ros2 node list` 确认 dnn_node 正在运行 |
| Web 无 AI 结果叠加 | websocket 未连接 AI topic | 确认 websocket 订阅的 topic 名称正确 |
| 帧率极低 (<5fps) | 模型过大或分辨率过高 | 降低输入分辨率;使用轻量模型(如 YOLOv5s 替代 YOLOv5x) |
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!