About | Support | answers.ros.org	Search:

Recognizing tools

Description: This tutorial explains how to recognize objects using tod_detecting

Tutorial Level:

Configuration

Tod works well with the following config.yaml :

%YAML:1.0
# TODParameters
TODParameters:
   # GuessParameters
   GuessParameters:
      min_cluster_size: 5
      min_inliers_count: 5
      ransac_iterations_count: 100
      max_projection_error: 6.
      descriptor_distance_threshold: 0.5
      min_stddev_factor: 0.0
   # FeatureExtractionParams
   feature_extraction_params:
      detector_type: DynamicFAST
      extractor_type: multi-scale
      descriptor_type: rBRIEF
      detector_params:
         max_features: 1500.
         min_features: 1000.
         threshold: 200.
      extractor_params:
         octaves: 3.
   # MatcherParameters
   MatcherParameters:
      matcher_type: LSH-BINARY
      ratio_threshold: 0.8
      knn: 3
      do_ratio_test: 1
   # ClusterParameters
   ClusterParameters:
      max_distance: 50

Recognizer sample

Tool for object recognizing on the one test image. Allowed options:

  --help                    Produce help message.
  -I [ --image ] arg        Query image. Required.
  -B [ --base ] arg (=./)   The directory that the training base is in.
  -f [ --tod_config ] arg   The name of the configuration file
  -V [ --verbose ] arg (=1) Verbosity level
  -m [ --mode ] arg (=0)    Mode: 0-textured object detection, 1 - kinect 
                            version
if verbose == 0 it doesn't draw any images
if verbose == 1 it draws projection
if verbose == 2 it draws projection and matches

if mode == 0, it uses TODRecognizer class (wrapper on algorithm from textured_object_detection package)
if mode == 1, it uses KinectRecognizer class (wrapper on algorithm from recognition_node application)

Example:

bin/recognizer -I /shared/test/t0004/images/25.png -B /shared/base/ -f /shared/config.yaml -V 1 -m 1

Recognizer sample for directory with test images

Tool for object recognizing on the one test image. Allowed options:

  --help                    Produce help message.
  -I [ --images ] arg       Folder with query images. Required.
  -B [ --base ] arg (=./)   The directory that the training base is in.
  -f [ --tod_config ] arg   The name of the configuration file
  -l [ --log ] arg          The name of the log file
  -V [ --verbose ] arg (=1) Verbosity level
  -m [ --mode ] arg (=0)    Mode: 0-textured object detection, 1 - kinect 
                            version
if verbose == 0 it doesn't draw any images
if verbose == 1 it draws projection
if verbose == 2 it draws projection and matches

if mode == 0, it uses TODRecognizer class (wrapper on algorithm from textured_object_detection package)
if mode == 1, it uses KinectRecognizer class (wrapper on algorithm from recognition_node application)

Log file used for calculation of detection rate in tod_test package. 

Example:

bin/frecognizer -I /shared/test/t0004/images -B /shared/base/ -f /shared/config.yaml -V 1 -m 0 -l log.xml