Forget Sharpness: Perturbed Forgetting of Model Biases Within SAM Dynamics 

[Code]

DynaShare: Task and Instance Conditioned Parameter Sharing for Multi-task Learning 

[Code] 

Meta Temporal Point Process

[Code] 

• Topometric Localization with Deep learning 

  • Paper Accepted for The International Symposium on Robotics Research (ISRR). Arxiv version available at: Arxiv_paper 

• Efficient Deep Models for Monocular Road Segmentation

  • We proposed a new architecture called Fast-Net for near real-time performance.

• Deep Learning for human part segmentation

•     We are investigating CNN architectures to perform human part segmentation. 

• Completed Projects 

  • View Planning For Cloud-Based Active Object Recognition

  •     We are investigating the possibility of enabling cloud-based object recognition by carefully planning the viewpoints.

• • Real Time Action Recognition

  •     We proposed the Space-time Occupancy Patterns, a new visual representation for 3D action recognition from sequences of depth maps which is the base for real time human robot interaction.

• • ROS TurtleBot Like construction

  •     This project developed a ROS TurtleBot Like robotic Plataform which we call Michelangelo. The video below show the first configuration obtained.

• • • Second video show our new configuration performing Autonomous navigation. The map was created using GMapping provide by ROS. Based on Willow Garage specification we built a power and gyro circuit to reach needed capabilities necessary to use ROS mapping and navigation nodes.

• • RGB-D Binary Descriptor  

  • Designed  an accurate and meaningful  maps of indoor environments based on objects. A sort of sensors can be employed for this task, such as, lasers and/or cameras, however with the recently introduction of fast and inexpensive RGB-D sensors (RGB actually means trichromatic intensity information and D depth). The Integration of synchronized intensity (color) and depth feasible. We focus on how represent geometrical information with a higher level of abstration.

  • Nascimento, E. ; OLIVEIRA, G. L. ; VIEIRA, A. W. ; Campos, M. . Improving Object Detection and Recognition for Semantic Mapping with an Extended Intensity and Shape based Descriptor. In: IROS 2011 workshop - Active Semantic Perception and Object Search in the Real World (ASP-AVS-11), 2011, San Francisco. Proc. IROS Workshop ASP-AVS-11, 2011. PDF 

• • Experiments with image reconstruction were performed to show the robustness of the method.

• • Underwater Visual SLAM

  • The use of Autonomous Underwater Vehicles (AUVs) for visual inspection tasks is a promising robotic field. The images captured by the robots can also aid in their localization/navigation. In this context, this project proposed an approach to localization and mapping problem of underwater vehicle. Supposing the use of inspection cameras, this proposal is composed of two stages: First the use of computer vision through the algorithm SIFT to extract the features in underwater image sequences and second the development of topological maps to localization and navigation. The integration of such systems will permit simultaneous localization and mapping of the environment. A set of tests with real robots was accomplished, regarding online and performance issues. The results reveals an accuracy and robust approach to several bottom conditions, illumination and noise, leading to a promissory and original SLAM technique.

  • Botelho S.S.C; OLIVEIRA, G. L. ; HAFFELE, C. ; Figueiredo M. ; Drews P. Self-Localization and Mapping for underwater Autonomous Vehicles. In: Edited by Hanafiah Yussof, In-Tech Press. (Org.). Robot Localization and Map Building. Vienna: In-Tech, 2009, v. 1, p. -.PDF 

• • • CATADIOPTRIC VISUAL ODOMETRY

    •  We propose a method for visual odometry using optical flow with a single omnidirectional (catadioptric) camera. We show how omnidirectional images can be used to perform optical flow, discussing the basis of optical flow and some restrictions needed to it and how unwarp these images. In special we describe how to unwarp omnidirectional images to Bird’s eye view, that correspond to scaled orthographic views of the ground plane. Catadioptric images facilitate landmark based odometry, since landmarks remain visible for longer time, as opposed to a small field-of-view standard camera. Also, providing adequate representations to support visual odometry with a fast processing time.

    • OLIVEIRA, G. L. ; Nascimento, E. ; Campos, M. Visual odometry with omnidirectional images In: Simposio Brasileiro de Automacao Inteligente (SBAI), 2011, Sao Joao del-Rei - MG. Anais Simpisio Brasileiro de Automacao Inteligente, 2011. PDF 

Final course work 

• Monocular Visual odometry, investigate aplications of monocular visual SLAM and test a technique  based on neural networks, called RATSLAM.

• Reducing planar rendezvous to 1D search.

• Human segmentation using stochastic background subtraction, based on ViBe algorithm.

Useful Links

• My Google Scholar

• Linkedin