We present a method for 3D object detection and pose estimation from single image. In contrast to current techniques that only regress the orientation of an object, our first regresses relatively stable properties using deep convolutional neural network then combines these estimates with geometric constraints provided by 2D bounding box produce complete box. The output novel hybrid discrete-continuous loss, which significantly outperforms L2 loss. second dimensions, have little variance...

Skillful mobile operation in three-dimensional environments is a primary topic of study Artificial Intelligence. The past two years have seen surge creative work on navigation. This output has produced plethora sometimes incompatible task definitions and evaluation protocols. To coordinate ongoing future research this area, we convened working group to empirical methodology navigation research. present document summarizes the consensus recommendations group. We discuss different problem...

In this paper we present a prototype system for image based localization in urban environments. Given database of views city street scenes tagged by GPS locations, the computes location novel query view. We first use wide-baseline matching technique on SIFT features to select closest database. Often due large change viewpoint and presence repetitive structures, percentage matches (> 50%) are not correct correspondences. The subsequent motion estimation between view reference view, is then...

What is a good visual representation for navigation? We study this question in the context of semantic navigation, which problem robot finding its way through previously unseen environment to target object, e.g. go refrigerator. Instead acquiring metric map an and using planning our approach learns navigation policies on top representations that capture spatial layout contextual cues. propose use segmentation detection masks as observations obtained by state-of-the-art computer vision...

We present a new public dataset with focus on simulating robotic vision tasks in everyday indoor environments using real imagery. The includes 20,000+ RGB-D images and 50,000+ 2D bounding boxes of object instances densely captured 9 unique scenes. train fast category detector for instance detection our data. Using the we show that, although increasingly accurate fast, state art is still severely impacted by scale, occlusion, viewing direction all which matter robotics applications. next...

Theoretical and analytical aspects of the visual servoing problem have not received much attention. Furthermore, estimation from vision measurements has been considered separately design control strategies. Instead addressing pose problems separately, we attempt to characterize types tasks which can be achieved using only quantities directly measurable in image, bypassing phase. We consider task navigation for a nonholonomic ground mobile base tracking an arbitrarily shaped continuous curve....

Detection of objects in cluttered indoor environments is one the key enabling functionalities for service robots. The best performing object detection approaches computer vision exploit deep Convolutional Neural Networks (CNN) to simultaneously detect and categorize interest scenes. Training such models typically requires large amounts annotated training data which time consuming costly obtain. In this work we explore ability using synthetically generated composite images state-of-the-art...

Multi-view stereo methods frequently fail to properly reconstruct 3D scene geometry if visible texture is sparse or the exhibits difficult self-occlusions. Time-of-Flight (ToF) depth sensors can provide information regardless of but with only limited resolution and accuracy. To find an optimal reconstruction, we propose integrated multi-view sensor fusion approach that combines from multiple color cameras ToF sensors. First, measurements are combined obtain a coarse complete model. Then,...

Surface electromyography (sEMG) has been the predominant method for sensing electrical activity a number of applications involving muscle-computer interfaces, including myoelectric control prostheses and rehabilitation robots. Ultrasound imaging mechanical deformation functional muscle compartments can overcome several limitations sEMG, inability to differentiate between deep contiguous compartments, low signal-to-noise ratio, lack robust graded signal. The objective this study was evaluate...

City environments often lack textured areas, contain repetitive structures, strong lighting changes and therefore are very difficult for standard 3D modeling pipelines. We present a novel unified framework creating city models which overcomes these difficulties by exploiting image segmentation cues as well presence of dominant scene orientations piecewise planar structures. Given panoramic street view sequences, we first demonstrate how to robustly estimate camera poses without need bundle...

Multi-scale deep CNNs have been used successfully for problems mapping each pixel to a label, such as depth estimation and semantic segmentation. It has also shown that architectures are reusable can be multiple tasks. These networks typically trained independently task by varying the output layer(s) training objective. In this work we present new model simultaneous segmentation from single RGB image. Our approach demonstrates feasibility of parts then fine tuning full, combined on both...

For applications in navigation and robotics, estimating the 3D pose of objects is as important detection. Many approaches to estimation rely on detecting or tracking parts keypoints [11, 21]. In this paper we build a recent state-of-the-art convolutional network for sliding-window detection [10] provide rough single shot, without intermediate stages initial bounding boxes. While not first system treat categorization problem, attempt combine at same level using deep learning approach. The key...

With the increasing speeds of modern microprocessors, it has become ever more common for computer-vision algorithms to find application in real-time control tasks. In this paper, we present an analysis problem steering autonomous vehicle along a highway based on images obtained from CCD camera mounted vehicle. We explore effects changing various important system parameters like velocity, look-ahead range vision sensor, and processing delay associated with perception systems. also results...

This paper presents a nonparametric approach to semantic parsing using small patches and simple gradient, color location features. We learn the relevance of individual feature channels at test time locally adaptive distance metric. To further improve accuracy approach, we examine importance retrieval set used compute nearest neighbours novel descriptor retrieve better candidates. The is validated by experiments on several datasets for demonstrating superiority method compared state art approaches.

Vision-based topological localization and mapping for autonomous robotic systems have received increased research interest in recent years. The need to map larger environments requires models at different levels of abstraction additional abilities deal with large amounts data efficiently. Most successful approaches appearance-based datasets typically represent locations using local image features. We study the feasibility performing these tasks urban global descriptors instead taking...

Multi-scale deep CNNs have been used successfully for problems mapping each pixel to a label, such as depth estimation and semantic segmentation. It has also shown that architectures are reusable can be multiple tasks. These networks typically trained independently task by varying the output layer(s) training objective. In this work we present new model simultaneous segmentation from single RGB image. Our approach demonstrates feasibility of parts then fine tuning full, combined on both...

Finding correspondences between images or 3D scans is at the heart of many computer vision and image retrieval applications often enabled by matching local keypoint descriptors. Various learning approaches have been applied in past to different stages pipeline, considering detection, description, metric objectives. These objectives were typically addressed separately most previous work has focused on data. This paper proposes an end-to-end framework for detection its representation...

This paper presents a new multi-view RGB-D dataset of nine kitchen scenes, each containing several objects in realistic cluttered environments including subset from the BigBird dataset. The viewpoints scenes are densely sampled and annotated with bounding boxes 3D point cloud. Also, an approach for detection recognition is presented, which comprised two parts: (i) proposal generation method (ii) development baselines using AlexNet to score our proposals, trained either on crops or...

Gestures in American Sign Language (ASL) are characterized by fast, highly articulate motion of upper body, including arm movements with complex hand shapes and facial expressions. In this work, we propose a new method for word-level sign recognition from using video. Our uses both shape cues while being robust to variations execution. We exploit the knowledge body pose, estimated an off-the-shelf pose estimator. Using as guide, pool spatio-temporal feature maps different layers 3D...