How gerunsj/deteccion-de-personas-con-turtlebot-y-opencv-1 works
Main Components
• Evaluation Modules – Found under the “evaluacion” directory. Here you see:
• “clasificadores” (various Haar cascades and LBP detectors)
• “src” (C++ source code for different detectors such as detector_final.cpp, detector_hog.cpp, etc.)
• “matlab” – scripts for additional evaluation, image analysis (e.g., histograma.m)
• “scripts” – shell scripts that coordinate running various evaluation executables and possibly batch processing sets of images.
• “resultados” – folders with CSV/ODS/text outputs, showing test runs over different datasets (set1, set235nuevo, mejores235, etc.)
• ROS Implementation – Under “implementacion” there is a ROS workspace (with catkin_make configuration) containing the “detpeople” package. This package includes:
• ROS node(s) that subscribe (and, in one case, publish video) on topics such as “/camera/image”
• Files like ImageConverter.cpp (used to convert image messages to OpenCV format), as well as various detector implementations (DetectorCascada, DetectorHog, etc.)
• Peripheral / Calibration Data – The “otros” directory contains calibration files for the TurtleBot’s cameras, and an external raspi node zip file for the Raspicam.
• Documentation – PDFs such as “Guía_de_uso.pdf” and the project “informe_Aguado_Emder.pdf” serve as background/support documents.
Relationships and Data Flow
• ROS Communication: Two key ROS nodes exist:
• The “publicador_video” node (in implementacion package) that publishes live image feeds.
• The “detpeople” detector node that subscribes to input topics, applies detection algorithms (which internally may use one or several classifiers such as Haar cascades or HOG detectors), and publishes detection results.
– Note the customization indicated in ImageConverter.cpp (you may switch between input sources).
• Evaluation Pipeline: In parallel to the live detection:
• The “evaluacion” components (compiled executables in evaluacion/ejecutables and source code in evaluacion/src) run offline experiments on different datasets (with results stored in evaluacion/resultados).
• Matlab scripts process images (see evaluacion/matlab) and help generate histograms or other analyses.
• External Dependencies: OpenCV (for processing, detection, image manipulation) and ROS Indigo (for node communication).
Architectural Patterns and Design Principles
• Modularity: The obvious separation between evaluation and online modules allows iterative testing without impacting the live system.
• ROS Pub/Sub Design: Making use of topics to decouple the image source and processing.
• Reusability: Several detectors are implemented in different ways (e.g., “detector_final”, “detector_hog”, “detector_dummy”) which suggests an architecture that can easily swap algorithms.
• Scripted Orchestration: Shell scripts in evaluacion/scripts organize and automate evaluation experiments.
• Separation of Concerns: The project differentiates between testing (offline evaluation, result generation) and operational modules (online detection with ROS).