2D/3D Object Recognition & Categorization
 qualitative_3c_small [2014] Interest points via Maximal Self-Dissimilarities

MSD is an interest point detector for images based on the intuition that image patches which are highly dissimilar from others over a relatively large extent of their surroundings are repeatable and distinctive.

[MSD Code] [Pdf (ACCV 14)]
 qualitative_3c_small [2013] BOLD features for texture-less object detection

BOLD features are image descriptors specifically conceived to boost performance on texture-less object detection. At the same time, they are efficiently computable and able to deal with textured object surfaces as well.

[Project Page] [Pdf (ICCV 13)]
 qualitative_3c_small [2013] RGBD-D object recognition via multiple features

This work aims at RGB-D object recognition within clutter and occlusion by means of Global Hypothesis Verification as well as multiple types of features (3D local, 3D global, 2D local).

[Pdf (ICRA 13)]
 qualitative_3c_small [2012] Hypothesis Verification for 3D Object Recognition

A robust Hypothesis Verification stage is demonstrated to dramatically improve the performance of 3D Object Recognition algorithms based on local and global features. The proposed method can work with high occlusion levels and was able to yield 100% recognition rate on a standard benchmark dataset for 3D Object Recognition.

[Project Page] [Pdf (ECCV 12)]
 12_keypoint_eval [2012] Performance evaluation of 3D keypoint detectors 

The field of keypoint detectors has reached enough maturity to motivate a study including an analysis and classification of current methodologies and an experimental evaluation on different application scenarios.

[Project page] [Pdf (3DIMPVT 11)]

[2011] Semantic Segmentation of 3D data based on invariant features, Markov Random Field and general purpose classifiers.

The proposed method has been tested on different scenarios (indoor/outdoor) and sensor data such as structured light (MS Kinect), stereo, LIDAR.

[Demo Video] [Urban Data Results] [Pdf (3DIMPVT 11)][Pdf (IROS 11)]
 trism [2010] 3D Implicit Shape Models

Extension of the Implicit Shape Models (ISM) to object categorization of 3D data (TRISM)
[Pdf (ACCV 10)]
[2010] Hough-based Correspondence Grouping for 3D Object Recognition

We present a way to divide a set of 3D point-to-point correspondences into object hypotheses (correspondence grouping) for the task of 3D Object Recognition and 6DOF Pose Estimation in clutter and occlusion. This method relies on a novel Hough Voting scheme built over a 3D domain instead than on a 6D, with benefits in terms of sparseness of the solution and efficiency.
[Demo Video] [Pdf] [Pdf (invited journal extension)]
[2010] Unique Shape Context (USC)

USC is a 3D descriptor that builds on and improves the performance of the “3D Shape Context” descriptor
[2010] SHOT 3D Descriptor

SHOT is a novel 3D keypoint descriptor for robust surface matching under clutter and occlusion (ECCV 2010). SHOT has been later extended to handle RGB-D data as a joint Texture+Shape descritpor (Color-SHOT)
[Project page] [Pdf] [Pdf (Color-SHOT)]
Stereo Correspondence
  • [2010] Stereo for robots: quantitative evaluation of efficient and low-memory dense stereo algorithms (ICARCV 2010)
  • [2008] “Fast-Aggregation”: a novel algorithm deploying segmentation and incremental techniques to perform near real-time and accurate dense stereo matching (ICPR 2008)
  • [2008] A classification and performance evaluation of cost aggregation strategies for stereo matching (CVPR 2008)
  • [2007] “Segment-support”: accurate dense local stereo matching using an aggregation strategy based on a variable support (PSIVT07)
  • [2007] “SO+borders” accurate stereo matching based on cost aggregation and a Scanline Optimization framework (ACCV07)
3D acquisition and reconstruction
  • [2010] A 3D reconstruction system based on improved Spacetime Stereo (ICARCV 2010)
  • [2009] 3Dreconstruction of scenes with moving objects/people using regularization methods, stereo matching and pattern projection (ICINCO09)
Video analysis for video-surveillance
  • [2010] Background subtraction based on polynomial models (ICIP10, VS10)
[Pdf (ICIP)] [Pdf (VS)]
  • [2009] Non-linear parametric Bayesian regression for background subtraction (MOTION 2009)
  • [2009] Video analysis performing removed/abandoned object detection with a multimodal sensor (a PIR sensor and a camera) on a low-power embedded architecture (AVSS09, ECCV-MF2SFA209)
  • [2008] A novel multi-view change detection approach used to perform automatic detection of vandal act events such as graffiti (ECCV-VS08).
  • [2008] Reliable access monitoring and singularization in interlocks (AVSS08).
  • [2008] Automatic graffiti detection deploying a Time-of-Flight (TOF) sensor (ACIVS08).
  • [2007] Change detection approach based on background subtraction robust to sudden illumination changes occurring in the scene. (WACCV07).
  • [2007] How to usefully deploy scene 3D information coming from a 3D device (e.g. a stereo vision system) in order to enhance robustness of a video-surveillance system (WACCV07).


Fast exhaustive pattern matching and block matching
  • [2012] Performance evaluation of Full-Search equivalent pattern matching algorithms [Pdf] [Project page with code, results and dataset]
  • [2012] Adaptive Low Resolution Pruning (ALRP): a Full-Search equivalent algorithm improving the computational efficiency of the state of the art [Pdf]
  • [2011] ZEBC: a generalization of EBC to the ZNCC measure and to multi-channel images [Pdf] [Pdf]
  • [2009] Incremental Dissimilarity Approximations (IDA): fast exhaustive pattern matching based on Lp norm-based measures (PAMI ’09) [Pdf]
  • [2008] Enhanced Bounded Correlation (EBC): an exhaustive technique to speed-up template matching based on Normalized Cross Correlation (TIP ’08) [Pdf]
  • [2007] Fast exhaustive block matching for motion estimation (ICIAP ’07) [Pdf]
  • [2005] Use of SIMD instructions to increase the efficiency of template matching methods based on correlation measures (CAMP05) [Pdf]
Robust visual correspondence
  • [2008] Performance evaluation of robust matching measures proposed in literature for the pattern matching task (VISAPP2008).
  • [2007] A class of matching measures for visual correspondence that are robust toward disturbance factors such as sudden illumination changes, noise, occlusions. They can be usefully adopted for tasks such as pattern matching, image registration, stereo vision, change detection (ICIAP07).