melodic   Show EOL distros: 

Description

The intermediate object generator generates intermediate object for a domain composed from scenes. It is used to restrain the amount of object searched during the direct search (asr_state_machine) phase by selecting appropriate objects to search.

Functionality

The intermediate objects are calculated by using 3 criteria :

1. The position variance of the object. The more an object moves, the less likely it is to be an intermediate object. To get this value, first we need to get the average position, and then calculate the distance between the average object qnd the seen object in every snapshot.

2. The average distance between the current tested object and others objects in the scene. The value is calculated with the average distance in each snapshot. Then we average the value between all snaposhots.

3. The presence of the object in the scene. This value is quite simple to calculate : It is number of snapshot where the object is present divided by the total number of snapshots.

The second part of the process is to normalize all values to hqve the same scale of comparison between all 3 criteria.

Then we apply the Evaluation Function F (either by addition or by multiplication ):

After that we can apply the following algorithm to filter the object list and only get the intermediate object.

At every step we calculate the average F value of the list qnd once the gain by removing an object is under a defined threshold, we stop the process. In addition, we have to test the validity of the list at each steps (Each scene is represented by at least one object). Please note that the list is sorted by F value before applying the Filter.

This package needs a database of objects and their absolute positions during the time (can be set in the asr_recognizer_prediction_ism). On basis of this information it will be determined which objects have a heigher chance to be found in a 3D-object-search. There will be two files saved.

In one a subset of the current objects will be saved, to limit the search objects while the direct search of the asr_state_machine.

In the second the calculated weight for each object will be saved. This can be used in the asr_next_best_view to prefer some objects about the others.

The weights will also be saved in a new table in the database.

Usage

he asr_recognizer_prediction_ism/launch/helpers/scene_recognition.yaml is needed to access the path to the current database.

Every scene has to have at least 3 objects or the program crashes.

Needed software

Boost (1.54.0.1)

sqlite3

Start system

Update the path to the database in asr_recognizer_prediction_ism/launch/helpers/scene_recognition.yaml

roslaunch asr_intermediate_object_generator asr_intermediate_object_generator.launch

The asr_world_model will launch this package automatically if the generated files for the current database were not generated yet. There is an option in asr_world_model to activate this function.

ROS Nodes

Parameters

Launch file

asr_intermediate_object_generator/launch/asr_intermediate_object_generator.launch

!AutomatPath: The path were the intermediate_objects xml file should be created. If there is an XXX in the name, the XXX will be replaced with the current database name. So the name has not to be changed manually after a database change.

NBVPath: The path were the intermediate_object_weights xml file should be created. The XXX works analog here, too.

yaml file

asr_intermediate_object_generator/param/Domain.yaml

!RankingMethod : 0 or 1, 0 is additive and 1 is multiplicative Alpha : weight for the presence in scene criteria Beta : weight for the position variance criteria Gamma : weight for the average distance between object criterias! GainValueThreshold : gain threshold for the object filtering !DomainName : name of the domain