Warning: fopen(misc/lockfile/1a025426414e31ca086ec6921df59c0c): failed to open stream: Permission denied in /home/pretto/public_html/shared.php on line 194 Warning: flock() expects parameter 1 to be resource, null given in /home/pretto/public_html/shared.php on line 230
 
Login


:



Research

Agriculture Robotics

Flourish Project

Flourish is a project of precision agriculture that aims to reduce the amount of herbicides used to control weeds by means of the use of robotics and artificial intelligence technologies. A flying robot (UAV) and an autonomous ground vehicle (UGV) cooperate in an almost fully automated way to monitor the crop and to precisely remove the weeds. The UAV is in charge to fly over and inspect the crops from the sky, the UGV uses the information provided by the UAV in order to reach and remove the detected weeds.
The Flourish project is funded by the European Community's Horizon 2020 programme.




Papers:

  • C. Potena, D. Nardi and A. Pretto Fast and Accurate Crop and Weed Identification with Summarized Train Sets for Precision Agriculture In Proceedings of 14th International Conference on Intelligent Autonomous Systems (IAS-14), July 3-7 , 2016 Shanghai, China (In press)
    (PDF, BibTeX)
    @inproceedings{pnp_ias2016,
      author={Potena, C. and Nardi, D. and Pretto, A.},
      title={Fast and Accurate Crop and Weed Identification
             with Summarized Train Sets for Precision Agriculture},
      booktitle={Proc. of 14th International Conference on 
                 Intelligent Autonomous Systems (IAS-14)},
      year={2016}
    }
    


    Object detection and Localization

    FlexSight Project

    The goal fo the FlexSight project (Flexible and Accurate Recognition and Localization System of Deformable Objects for Pick&Place Robots) is to design a perception system based on an integrated smart camera (the FlexSight sensor, FSS) that is able to recognize and localize several types of deformable objects that can be commonly found in many industrial and logistic applications. We refer to a "deformable object" either as an object that can change its shape due to a stress, or that can be obtained by applying a resize operator along one or more directions.


    The FSS sensor will integrate all the required sensors (depth sensor, stereo RGB camera, ...) and processing units (CPU+GPU) suitable to run the implemented algorithms and it will be one of the first smart camera that explicitly deals with deformable objects: our system will search over a parameters set that not only includes the class and the position of the object, but also its deformation/rescaling parameters.
    The FlexSight project is funded by the European Community's project ECHORD++.


    Here a video of the demo of an early version of the FlexSight sensor prototype presented at Hannover Messe 2017:


    Active Textureless Object Recognition and Localization




    We propose an active object detection and localization framework that combines a robust untextured object detection and 3D pose estimation algorithm with a novel next-best-view selection strategy.
    We address the detection and localization problems by proposing an edge-based registration algorithm (D2CO) that refines the object position by minimizing a cost directly extracted from a 3D image tensor (Directional Chamfer Distance, DCD) that encodes the minimum distance to an edge point in a joint direction/location space. Unlike the Chamfer distance, the DCD takes into account also the edges directions.



    We extract a set of object candidates by pre-computing the (projected) raster templates along with their image orientations for a large number of possible 3D locations; for each point, we lookup the DCD tensor, computing the template average distance. The templates are then sorted for increasing distances. D2CO finally refines the object position employing a non-linear optimization procedure that minimizes a tensor-based cost function.





    We face the next-best-view problem by exploiting a sequential decision process that, for each step, selects the next camera position which maximizes the mutual information between the state and the next observations. We solve the intrinsic intractability of this solution by generating observations that represent scene realizations, i.e. combination samples of object hypothesis provided by the object detector, while modeling the state by means of a set of constantly resampled particles.





    D2CO webpage with code

    Papers:

  • M. Imperoli and A. Pretto D2CO: Fast and Robust Registration of 3D Textureless Objects Using the Directional Chamfer Distance In Proceedings of the 10th International Conference on Computer Vision Systems (ICVS 2015), July 6-9 , 2015 Copenhagen, Denmark, pages: 316-328
    (PDF, BibTeX)
    @inproceedings{miap_icvs2015,
      author={Imperoli, M. and Pretto, A.},
      title={{D\textsuperscript{2}CO}: Fast and Robust Registration of {3D}
             Textureless Objects Using the {Directional 
             Chamfer Distance}},
      booktitle={Proc. of 10th International Conference on 
                 Computer Vision Systems (ICVS 2015)},
      year={2015},
      pages={316--328}
    }
    


  • M. Imperoli and A. Pretto Active Detection and Localization of Textureless Objects in Cluttered Environments In arXiv preprint arXiv:1603.07022
    (PDF, BibTeX)
    @article{miap_arxiv2016,
      title={Active Detection and Localization of Textureless Objects 
             in Cluttered Environments},
      author={Imperoli, Marco and Pretto, Alberto},
      journal={arXiv preprint arXiv:1603.07022},
      year={2016}
    }
    


    Industrial Bin-Picking with Monocamera Vision System


    We present a robust and flexible vision system for 3D localization of planar parts for industrial robots. Our system is able to work with nearly any object with planar shape, randomly placed inside a standard industrial bin or on a conveyor belt. Our solution using a 2D camera provides to be a reliable, cost-effective and less invasive system to be installed in existing robotic cells. The main contributions of this work are:
    - An effective iterative optimize-and-score registration procedure, that employs a constrained optimization strategy that avoids the registration from being stuck on local minima. Actually, these minima often represent unreal objects created by the not perfect superimposition of two objects;
    - A smooth cost function based on a dynamically adapted gradient magnitude.



    Papers:

  • A. Pretto, S. Tonello, E. Menegatti Flexible 3D Localization of Planar Objects for Industrial Bin-Picking with Monocamera Vision System In IEEE International Conference on Automation Science and Engineering (IEEE CASE 2013), Madison, Wisconsin, (USA), August 17-21, 2013
    (PDF, BibTeX)
    @inproceedings{prettoCASE2013,
      title={Flexible 3D Localization of Planar Objects for Industrial 
    	 Bin-Picking with Monocamera Vision System},
      author={Pretto, A. and Tonello, S. and Menegatti, E.},
      booktitle={Proc. of: IEEE International Conference on 
               Automation Science and Engineering (CASE)},
      year={2013},
      pages={168 -- 175}
    }
    


    Visual SLAM

    Visual-Inertial Ego-Motion Estimation


    We describe an ego-motion estimation system developed specifically for humanoid robots, integrating visual and inertial sensors. It addresses the challenge of significant scale changes due to forward motion with a finite field of view by using recent sparse multi-scale feature tracking techniques. Additionally, it addresses the challenge of long-range temporal correlation due to walking gaits by employing a kinematic-statistical model that does not require accurate knowledge of the robot dynamics and calibration.



    The walking motion experiment is a significantly challenging dataset comprised of brisk walking using the head-mounted system through the hallways of a rectangular building. Throughout the dataset the operator frequently changes gaze direction. On this challenging example of humanoid legged motion we achieve a drift of 1.01% over a 292m course.



  • K. Tsotsos, A. Pretto, S. Soatto Visual-Inertial Ego-Motion Estimation for Humanoid Platforms Proc. of the IEEE-RAS International Conference on Humanoid Robots (Humanoids 2012), Osaka (Japan), Pages: 704-711 (PDF, BibTeX)
    @inproceedings{tps_humanoids2012,
      title={Visual-Inertial Ego-Motion Estimation for Humanoid Platforms},
      author={Tsotsos, K. and Pretto, A. and Soatto, S.},
      booktitle={IEEE-RAS International Conference on Humanoid Robots},
      year={2012},
      pages={704--711}
    }
    


    3D Dense Structure Reconstruction Based on Meaningful Triangulation


    In this work, we propose an effective dense 3D reconstruction and navigation system based on omnidirectional vision well suited for large scale scenarios. Aim of this work is to provide a robust and efficient way to build 3D maps with exhaustive information about both the structure and the appearance of the environment.
    We start from the assumption that the surrounding environment (the scene) forms a piecewise smooth surface represented by a triangle mesh. Our system is able to infer the structure of the environment along with the ego-motion of the camera performing a robust tracking of the projection in the omnidirectional image of this surface.





    Delaunay triangulation based on corner features is used as an initial guess of the surface tessellation; the triangulation is hence modified using a constrained edge insertion algorithm, where the edgelet features detected in the image represent the new predefined edges which are added to the new triangulation. Both the camera motion and the 3D environment structure parameters are estimated using a direct method inside an optimization framework, taking into account the topology of the subdivision in a robust and efficient way.





    Papers:

  • A. Pretto, E. Menegatti and E. Pagello Omnidirectional Dense Large-Scale Mapping and Navigation Based on Meaningful Triangulation In: In: Proceedings of the 2011 IEEE International Conference on Robotics and Automation (ICRA 2011), May 9-13, 2011 Shanghai (China), Pages: 3289 - 3296
    (PDF, BibTeX)
    @inproceedings{prettoICRA2011,
      author    = {Pretto, A. and Menegatti, E. and
                   Pagello, E.},
      title     = {Omnidirectional Dense Large-Scale Mapping and Navigation 
                   Based on Meaningful Triangulation},
      booktitle = {Proc. of. IEEE International Conference 
                   on Robotics and Automation (ICRA)},
      year      = {2011},
      pages     = {3289--3296},
    }
    


  • A. Pretto, S. Soatto and E. Menegatti Scalable Dense Large-Scale Mapping and Navigation In: 2010 IEEE International Conference on Robotics and Automation, Workshop on Omnidirectional Robot Vision.
    (PDF, BibTeX)
    @INPROCEEDINGS{prettoOVR2010,
    AUTHOR = {Pretto, A. and Soatto, S. and Menegatti, E.},
    TITLE = {Scalable Dense Large-Scale Mapping and Navigation},
    BOOKTITLE = {Proc. of:  Workshop on Omnidirectional 
                 Robot Vision (ICRA)},
    YEAR = {2010}
    }
    


    Go to the Old Projects section



    print

  •  
    Latest News

  • June, 28 2017: Our paper "Effective Target Aware Visual Navigation for UAVs" has been accepted at ECMR 2017

  • June, 15 2017: Our paper "Automatic Model Based Dataset Generation for Fast and Accurate Crop and Weeds Detection" has been accepted at IROS 2017

  • April 2017: We successfully presented a demo of an early version of the FlexSight sensor prototype at Hannover Messe 2017: see a video in the ECHORD++ FlexSight Project section.

  • July 2016: Our paper entitled "Fast and Accurate Crop and Weed Identification with Summarized Train Sets for Precision Agriculture" has been selected as one of the finalists for the 14th International Conference on Intelligent Autonomous Systems (IAS-14) Best Student Paper Award!