瑞莎星睿 O6 (Radxa Orion O6)-ubuntu24.04-ROS2 运行深度估计模型

引言

由Radxa联合此芯科技与安谋科技打造的"星睿O6"迷你ITX主板堪称当前最受期待的开发板之一。该产品搭载的CIX P1（CD8180）12核Armv9处理器配合30TOPS算力的NPU和高性能GPU，结合最高64GB LPDDR内存，非常适合AI开发工作站、边缘计算节点、以及高性能个人计算应用。

在这篇文章中，我将尝试在"星睿O6"上通过ROS2部署depth-anything-v2模型。

硬件要求

• "星睿O6" AI PC开发板
• USB摄像头

软件要求

• Ubuntu 24.04系统
• 安装ROS2

ROS2部署depth-anything-v2模型步骤

创建USB摄像头功能包和图像发布节点

第一步、创建功能包

mkdir -p ros_ws/src
cd ros_ws/src
ros2 pkg create image_publish --build-type ament_python --dependencies rclpy cv_bridge

第二步、编写发布节点代码

import rclpy
from rclpy.node import Node
from sensor_msgs.msg import Image
from cv_bridge import CvBridge
import cv2

class CameraPublisher(Node):
    def __init__(self):
        super().__init__('camera_publisher')
        self.publisher_ = self.create_publisher(Image, 'camera/image_raw', 10)
        self.cap = cv2.VideoCapture(0)
        self.bridge = CvBridge()
        self.timer = self.create_timer(0.1, self.timer_callback)

    def timer_callback(self):
        ret, frame = self.cap.read()
        if ret:
            msg = self.bridge.cv2_to_imgmsg(frame, "bgr8")
            self.publisher_.publish(msg)

def main():
    rclpy.init()
    node = CameraPublisher()
    rclpy.spin(node)
    node.destroy_node()
    rclpy.shutdown()

if __name__ == '__main__':
    main()

第三步、编译和执行

colcon build
source install/setup.bash
ros2 run image_publish image

第四步、查看话题

ros2 topic list

看见camera/image_raw话题说明USB摄像数据已经成功封装成ROS2数据格式并发布出来。

创建深度估计功能包和推理结果发布节点

第一步、创建功能包

cd ros_ws/src
ros2 pkg create depth_estimate --build-type ament_python --dependencies rclpy cv_bridge opencv-python numpy

第二步、编写推理节点代码
从CIX modelhub将depth_anything_v2预编译的.cix和NOE_Engine.py复制到ros_ws/src/depth_estimate/depth_estimate文件夹下。

import rclpy
from rclpy.node import Node
from sensor_msgs.msg import Image
from cv_bridge import CvBridge
import cv2
import numpy as np
from .NOE_Engine import EngineInfer

def preprocess_image(image,  mean : list = [0.485, 0.456, 0.406], std : list = [0.229, 0.224, 0.225], target_size : tuple = (520, 520), flag : bool = True, rgb : bool = True) -> np.ndarray:
    mean = np.array(mean).astype(np.float32)
    std = np.array(std).astype(np.float32)
    # image = cv2.imread(image_path)
    if rgb:
        image = image[:, :, ::-1]  # BGR2RGB
    image_resized = cv2.resize(image, target_size)
    if flag:
        image_normalized = image_resized.astype(np.float32) / 255.0
    else:
        image_normalized = image_resized.astype(np.float32)
    image_standardized = (image_normalized - mean) / std
    image_transposed = image_standardized.transpose(2, 0, 1)
    input_tensor = np.expand_dims(image_transposed, axis=0)
    return input_tensor

class CameraProcessor(Node):
    def __init__(self):
        super().__init__('image_processor')
        self.subscription = self.create_subscription(
            Image,
            'camera/image_raw',
            self.listener_callback,
            10)
        self.bridge = CvBridge()
        self.publisher_ = self.create_publisher(Image, 'processer/image_prcocess', 10)
        self.model = EngineInfer("/home/cix/ros_ws/src/depth_estimate/depth_estimate/depth_anything_v2.cix")

    def listener_callback(self, msg):
        input_image = self.bridge.imgmsg_to_cv2(msg, desired_encoding='bgr8')
        image = input_image
        h, w = image.shape[:2]
        input = preprocess_image(input_image,[0.485, 0.456, 0.406],[0.229, 0.224, 0.225],(518,518))
        depth = self.model.forward([input])[0]
        depth = np.reshape(depth, (1, 518, 518))
        depth = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
        depth = depth.transpose(1, 2, 0).astype("uint8")
        depth = cv2.resize(depth, (w, h), interpolation=cv2.INTER_CUBIC)
        depth_color = cv2.applyColorMap(depth, cv2.COLORMAP_INFERNO)

        ros_image = self.bridge.cv2_to_imgmsg(depth_color, encoding="bgr8")
        ros_image.header.stamp = self.get_clock().now().to_msg()
        ros_image.header.frame_id = "camera_frame"
        self.publisher_.publish(ros_image)

def main(args=None):
    rclpy.init(args=args)

    camera_processor = CameraProcessor()

    rclpy.spin(camera_processor)

    camera_processor.destroy_node()
    rclpy.shutdown()

if __name__ == '__main__':
    main()

第三步、编译和执行

colcon build
source install/setup.bash
ros2 run depth_estimate depth

查看话题

ros2 topic list

可以看到原始数据和深度估计的话题。
第四步、rviz可视化结果
在rviz订阅相关话题数据即可

ros2 run rviz2 rviz2

结语

总体来说，"星睿O6"在ROS2环境中，能够充分调用CPU和NPU的算了完成AI模型的推理，并通过ROS2的可视化插件展示出来。