Introduction#
After participating in the ROSConJP2025 workshop (Practical Introduction to ROS 2 and Model-Based Design (MBD)), I was able to enjoy ROS development with Simulink! So, this time I will try object recognition.
Caution#
In this article, detailed operations related to collaboration between MATLAB/Simulink and ROS are omitted.
Configuration#
- MacbookAir M3 (16GB)
- start matlab/simulink
- Camera data subscription and object recognition
- RaspberryPi4B (8GB)
- Publish camera data
Raspberry Pi side#
Required package installation#
sudo apt update
sudo apt install raspi-config v4l-util
Change settings#
sudo raspi-config
Select [3 Interface Options], enable [I1 Legacy Camera] and restart.
Starting camera node#
sudo apt install ros-jazzy-v4l2-camera
ros2 run v4l2_camera v4l2_camera_node --ros-args -p video_device:=/dev/video0
MATLAB/Simulink side#
detector=yolov4ObjectDetector("tiny-yolov4-coco");
save('detector.mat', 'detector');The function that filters only human detection is defined as below.
function [bboxes_person, isDetected] = filter_detections(boundingBoxes, labels)
% 検出された物体の数を取得
numDetections = size(labels, 1);
% personかどうかを格納するための論理配列を初期化
isPerson = false(numDetections, 1);
% forループを使って、ラベルを一つずつチェックする
for i = 1:numDetections
% ラベルを一つだけ取り出して文字列に変換し、"person"と比較
if string(labels(i)) == "person_label"
isPerson(i) = true;
end
end
% 1つでもpersonが検出されたかチェック
if any(isPerson)
% personのバウンディングボックスのみを抽出
bboxes_person = boundingBoxes(isPerson, :);
isDetected = true;
else
% personが検出されなかった場合、空行列を返す
bboxes_person = zeros(0, 4, 'single');
isDetected = false;
end
end