justin

Today I read a paper titled “Look-ahead before you leap: end-to-end active recognition by forecasting the effect of motion”

My initial thoughts: A lot of vision recognition systems, especially the simpler ones, are based on static imagery. If recognition of a moving scene is deployed it is rarely predictive and rarely (if ever) takes in to account the motion of the observer. Actually having demonstrable theories about how to deploy vision recognition in an observer system that is itself moving is hugely beneficial in all sorts of field applications.

Update: I’ve read stuff about this previously, in some other papers. Hmmm… need to go back to the cited works to see who these guys are referencing when I get home this evening.

The abstract is:
Visual recognition systems mounted on autonomous moving agents face the challenge of unconstrained data, but simultaneously have the opportunity to improve their performance by moving to acquire new views of test data.

In this work, we first show how a recurrent neural network-based system may be trained to perform end-to-end learning of motion policies suited for the “active recognition” setting.

Further, we hypothesize that active vision requires an agent to have the capacity to reason about the effects of its motions on its view of the world.

To verify this hypothesis, we attempt to induce this capacity in our active recognition pipeline, by simultaneously learning to forecast the effects of the agent’s motions on its internal representation of its cumulative knowledge obtained from all past views.

Results across two challenging datasets confirm both that our end-to-end system successfully learns meaningful policies for active recognition, and that “learning to look ahead” further boosts recognition performance.

Today I read a paper titled “Creating Simplified 3D Models with High Quality Textures”

Initial thoughts: More Kinect research. It’s a fun little toy, and hooking up multiple Kinects together gives me a full 360 degree scanning system with very little occlusion. Have been integrating the Kinect scanners in to some Hololens AR experiments where I process the Kinect data off-board on a workstation, then feed that to the Hololens via a dedicated data server so that I can overlay a transformational image on to the user’s body. Think of it like being able to see yourself wearing an IronMan suit, or that scene in BladeRunner 2049 where the digital waifu overlays herself with the human sex worker for a weird Replicant, hologram, human three way.

Combining the ideas from this paper, with the observer motion predictive ideas from another Kinect paper I am working through opens up a huge opportunity to “body match” the hologram that the HoloLens is displaying with the human user. It is going to take some work, but I think I am on to something here.

This paper is sort of tangentially related because what I am trying to do is scan (without a lot of noise) the user, get high resolution, non-occluded data, process it on a “lots of cores” workstation with a couple of high-end GPUs, and then feed that pre-processed model data in to the HoloLens for real-time holographic overlay and real-time model distortion of the user. Basically, I am making a “digital fat suit/fun house mirror” that mimics the human as they move around.

The abstract is:
This paper presents an extension to the KinectFusion algorithm which allows creating simplified 3D models with high quality RGB textures.

This is achieved through (i) creating model textures using images from an HD RGB camera that is calibrated with Kinect depth camera, (ii) using a modified scheme to update model textures in an asymmetrical colour volume that contains a higher number of voxels than that of the geometry volume, (iii) simplifying dense polygon mesh model using quadric-based mesh decimation algorithm, and (iv) creating and mapping 2D textures to every polygon in the output 3D model.

The proposed method is implemented in real-time by means of GPU parallel processing.

Visualization via ray casting of both geometry and colour volumes provides users with a real-time feedback of the currently scanned 3D model.

Experimental results show that the proposed method is capable of keeping the model texture quality even for a heavily decimated model and that, when reconstructing small objects, photorealistic RGB textures can still be reconstructed.

Today I read a paper titled “Noise in Structured-Light Stereo Depth Cameras: Modeling and its Applications”

My initial thoughts: I’ve done extensive experiments with the Kinect and one of the things that has always annoyed me is just how darn noisy the RGBD image data is coming off the device. I’ve developed some algorithms that denoise it, but a lot of how well your software works is completely dependent on environmental issues. Light bloom from an outside window can totally throw off your application and render t all but useless during a critical demo (trust me, I know this first hand).

I think I can make use of some of the denoising techniques covered in this paper, combined with a 3D planes segmentation algorithm.

The abstract is:
Depth maps obtained from commercially available structured-light stereo based depth cameras, such as the Kinect, are easy to use but are affected by significant amounts of noise.

This paper is devoted to a study of the intrinsic noise characteristics of such depth maps, i.e.

the standard deviation of noise in estimated depth varies quadratically with the distance of the object from the depth camera.

We validate this theoretical model against empirical observations and demonstrate the utility of this noise model in three popular applications: depth map denoising, volumetric scan merging for 3D modeling, and identification of 3D planes in depth maps.

Today I read a paper titled “Efficient Upsampling of Natural Images”

The abstract is:
We propose a novel method of efficient upsampling of a single natural image.

Current methods for image upsampling tend to produce high-resolution images with either blurry salient edges, or loss of fine textural detail, or spurious noise artifacts.

In our method, we mitigate these effects by modeling the input image as a sum of edge and detail layers, operating upon these layers separately, and merging the upscaled results in an automatic fashion.

We formulate the upsampled output image as the solution to a non-convex energy minimization problem, and propose an algorithm to obtain a tractable approximate solution.

Our algorithm comprises two main stages.

1) For the edge layer, we use a nonparametric approach by constructing a dictionary of patches from a given image, and synthesize edge regions in a higher-resolution version of the image.

2) For the detail layer, we use a global parametric texture enhancement approach to synthesize detail regions across the image.

We demonstrate that our method is able to accurately reproduce sharp edges as well as synthesize photorealistic textures, while avoiding common artifacts such as ringing and haloing.

In addition, our method involves no training phase or estimation of model parameters, and is easily parallelizable.

We demonstrate the utility of our method on a number of challenging standard test photos.

Today I read a paper titled “Universal Coating for Programmable Matter”

The abstract is:
The idea behind universal coating is to have a thin layer of a specific substance covering an object of any shape so that one can measure a certain condition (like temperature or cracks) at any spot on the surface of the object without requiring direct access to that spot.

We study the universal coating problem in the context of self-organizing programmable matter consisting of simple computational elements, called particles, that can establish and release bonds and can actively move in a self-organized way.

Based on that matter, we present a worst-case work-optimal universal coating algorithm that uniformly coats any object of arbitrary shape and size that allows a uniform coating.

Our particles are anonymous, do not have any global information, have constant-size memory, and utilize only local interactions.

Today I read a paper titled “Evolving Shepherding Behavior with Genetic Programming Algorithms”

The abstract is:
We apply genetic programming techniques to the `shepherding’ problem, in which a group of one type of animal (sheep dogs) attempts to control the movements of a second group of animals (sheep) obeying flocking behavior.

Our genetic programming algorithm evolves an expression tree that governs the movements of each dog.

The operands of the tree are hand-selected features of the simulation environment that may allow the dogs to herd the sheep effectively.

The algorithm uses tournament-style selection, crossover reproduction, and a point mutation.

We find that the evolved solutions generalize well and outperform a (naive) human-designed algorithm.

Today I read a paper titled “Artificial Persuasion in Pedagogical Games”

The abstract is:
A Persuasive Teachable Agent (PTA) is a special type of Teachable Agent which incorporates a persuasion theory in order to provide persuasive and more personalized feedback to the student.

By employing the persuasion techniques, the PTA seeks to maintain the student in a high motivation and high ability state in which he or she has higher cognitive ability and his or her changes in attitudes are more persistent.

However, the existing model of the PTA still has a few limitations.

Firstly, the existing PTA model focuses on modelling the PTA’s ability to persuade, while does not model its ability to be taught by the student and to practice the knowledge it has learnt.

Secondly, the quantitative model for computational processes in the PTA has low reusability.

Thirdly, there is still a gap between theoretical models and practical implementation of the PTA.

To address these three limitations, this book proposes an improved agent model which follows a goal-oriented approach and models the PTA in its totality by integrating the Persuasion Reasoning of the PTA with the Teachability Reasoning and the Practicability Reasoning.

The project also proposes a more abstract and generalized quantitative model for the computations in the PTA.

With higher level of abstraction, the reusability of the quantitative model is also improved.

New system architecture is introduced to bridge the gap between theoretical models and implementation of the PTA.

Today I read a paper titled “Query-Efficient Imitation Learning for End-to-End Autonomous Driving”

The abstract is:
One way to approach end-to-end autonomous driving is to learn a policy function that maps from a sensory input, such as an image frame from a front-facing camera, to a driving action, by imitating an expert driver, or a reference policy.

This can be done by supervised learning, where a policy function is tuned to minimize the difference between the predicted and ground-truth actions.

A policy function trained in this way however is known to suffer from unexpected behaviours due to the mismatch between the states reachable by the reference policy and trained policy functions.

More advanced algorithms for imitation learning, such as DAgger, addresses this issue by iteratively collecting training examples from both reference and trained policies.

These algorithms often requires a large number of queries to a reference policy, which is undesirable as the reference policy is often expensive.

In this paper, we propose an extension of the DAgger, called SafeDAgger, that is query-efficient and more suitable for end-to-end autonomous driving.

We evaluate the proposed SafeDAgger in a car racing simulator and show that it indeed requires less queries to a reference policy.

We observe a significant speed up in convergence, which we conjecture to be due to the effect of automated curriculum learning.

Today I read a paper titled “Dynamic Bayesian Networks to simulate occupant behaviours in office buildings related to indoor air quality”

The abstract is:
This paper proposes a new general approach based on Bayesian networks to model the human behaviour.

This approach represents human behaviour with probabilistic cause-effect relations based on knowledge, but also with conditional probabilities coming either from knowledge or deduced from observations.

This approach has been applied to the co-simulation of the CO2 concentration in an office coupled with human behaviour.

Today I read a paper titled “Egocentric Field-of-View Localization Using First-Person Point-of-View Devices”

The abstract is:
We present a technique that uses images, videos and sensor data taken from first-person point-of-view devices to perform egocentric field-of-view (FOV) localization.

We define egocentric FOV localization as capturing the visual information from a person’s field-of-view in a given environment and transferring this information onto a reference corpus of images and videos of the same space, hence determining what a person is attending to.

Our method matches images and video taken from the first-person perspective with the reference corpus and refines the results using the first-person’s head orientation information obtained using the device sensors.

We demonstrate single and multi-user egocentric FOV localization in different indoor and outdoor environments with applications in augmented reality, event understanding and studying social interactions.

Today I read a paper titled “The Computational Power of Dynamic Bayesian Networks”

The abstract is:
This paper considers the computational power of constant size, dynamic Bayesian networks.

Although discrete dynamic Bayesian networks are no more powerful than hidden Markov models, dynamic Bayesian networks with continuous random variables and discrete children of continuous parents are capable of performing Turing-complete computation.

With modified versions of existing algorithms for belief propagation, such a simulation can be carried out in real time.

This result suggests that dynamic Bayesian networks may be more powerful than previously considered.

Relationships to causal models and recurrent neural networks are also discussed.

Today I read a paper titled “Automatic Measurement of Physical Mobility in Get-Up-and-Go Test Using Kinect Sensor”

The abstract is:
Get-Up-and-Go Test is commonly used for assessing the physical mobility of the elderly by physicians.

This paper presents a method for automatic analysis and classification of human gait in the Get-Up-and-Go Test using a Microsoft Kinect sensor.

Two types of features are automatically extracted from the human skeleton data provided by the Kinect sensor.

The first type of feature is related to the human gait (e.g., number of steps, step duration, and turning duration); whereas the other one describes the anatomical configuration (e.g., knee angles, leg angle, and distance between elbows).

These features characterize the degree of human physical mobility.

State-of-the-art machine learning algorithms (i.e.

Bag of Words and Support Vector Machines) are used to classify the severity of gaits in 12 subjects with ages ranging between 65 and 90 enrolled in a pilot study.

Our experimental results show that these features can discriminate between patients who have a high risk for falling and patients with a lower fall risk.

Today I read a paper titled “Obstacle evasion using fuzzy logic in a sliding blades problem environment”

The abstract is:
This paper discusses obstacle avoidance using fuzzy logic and shortest path algorithm.

This paper also introduces the sliding blades problem and illustrates how a drone can navigate itself through the swinging blade obstacles while tracing a semi-optimal path and also maintaining constant velocity .

Today I read a paper titled “Remote Health Coaching System and Human Motion Data Analysis for Physical Therapy with Microsoft Kinect”

The abstract is:
This paper summarizes the recent progress we have made for the computer vision technologies in physical therapy with the accessible and affordable devices.

We first introduce the remote health coaching system we build with Microsoft Kinect.

Since the motion data captured by Kinect is noisy, we investigate the data accuracy of Kinect with respect to the high accuracy motion capture system.

We also propose an outlier data removal algorithm based on the data distribution.

In order to generate the kinematic parameter from the noisy data captured by Kinect, we propose a kinematic filtering algorithm based on Unscented Kalman Filter and the kinematic model of human skeleton.

The proposed algorithm can obtain smooth kinematic parameter with reduced noise compared to the kinematic parameter generated from the raw motion data from Kinect.

Today I read a paper titled “On the Computation of the Optimal Connecting Points in Road Networks”

The abstract is:
In this paper we consider a set of travelers, starting from likely different locations towards a common destination within a road network, and propose solutions to find the optimal connecting points for them.

A connecting point is a vertex of the network where a subset of the travelers meet and continue traveling together towards the next connecting point or the destination.

The notion of optimality is with regard to a given aggregated travel cost, e.g., travel distance or shared fuel cost.

This problem by itself is new and we make it even more interesting (and complex) by considering affinity factors among the users, i.e., how much a user likes to travel together with another one.

This plays a fundamental role in determining where the connecting points are and how subsets of travelers are formed.

We propose three methods for addressing this problem, one that relies on a fast and greedy approach that finds a sub-optimal solution, and two others that yield globally optimal solution.

We evaluate all proposed approaches through experiments, where collections of real datasets are used to assess the trade-offs, behavior and characteristics of each method.

Today I read a paper titled “Genetic cellular neural networks for generating three-dimensional geometry”

The abstract is:
There are a number of ways to procedurally generate interesting three-dimensional shapes, and a method where a cellular neural network is combined with a mesh growth algorithm is presented here.

The aim is to create a shape from a genetic code in such a way that a crude search can find interesting shapes.

Identical neural networks are placed at each vertex of a mesh which can communicate with neural networks on neighboring vertices.

The output of the neural networks determine how the mesh grows, allowing interesting shapes to be produced emergently, mimicking some of the complexity of biological organism development.

Since the neural networks’ parameters can be freely mutated, the approach is amenable for use in a genetic algorithm.

Today I read a paper titled “Recovering 6D Object Pose and Predicting Next-Best-View in the Crowd”

The abstract is:
Object detection and 6D pose estimation in the crowd (scenes with multiple object instances, severe foreground occlusions and background distractors), has become an important problem in many rapidly evolving technological areas such as robotics and augmented reality.

Single shot-based 6D pose estimators with manually designed features are still unable to tackle the above challenges, motivating the research towards unsupervised feature learning and next-best-view estimation.

In this work, we present a complete framework for both single shot-based 6D object pose estimation and next-best-view prediction based on Hough Forests, the state of the art object pose estimator that performs classification and regression jointly.

Rather than using manually designed features we a) propose an unsupervised feature learnt from depth-invariant patches using a Sparse Autoencoder and b) offer an extensive evaluation of various state of the art features.

Furthermore, taking advantage of the clustering performed in the leaf nodes of Hough Forests, we learn to estimate the reduction of uncertainty in other views, formulating the problem of selecting the next-best-view.

To further improve pose estimation, we propose an improved joint registration and hypotheses verification module as a final refinement step to reject false detections.

We provide two additional challenging datasets inspired from realistic scenarios to extensively evaluate the state of the art and our framework.

One is related to domestic environments and the other depicts a bin-picking scenario mostly found in industrial settings.

We show that our framework significantly outperforms state of the art both on public and on our datasets.

Today I read a paper titled “High statistics measurements of pedestrian dynamics”

The abstract is:
Understanding the complex behavior of pedestrians walking in crowds is a challenge for both science and technology.

In particular, obtaining reliable models for crowd dynamics, capable of exhibiting qualitatively and quantitatively the observed emergent features of pedestrian flows, may have a remarkable impact for matters as security, comfort and structural serviceability.

Aiming at a quantitative understanding of basic aspects of pedestrian dynamics, extensive and high-accuracy measurements of pedestrian trajectories have been performed.

More than 100.000 real-life, time-resolved trajectories of people walking along a trafficked corridor in a building of the Eindhoven University of Technology, The Netherlands, have been recorded.

A measurement strategy based on Microsoft Kinect\texttrademark has been used; the trajectories of pedestrians have been analyzed as ensemble data.

The main result consists of a statistical descriptions of pedestrian characteristic kinematic quantities such as positions and fundamental diagrams, possibly conditioned to local crowding status (e.g., one or more pedestrian(s) walking, presence of co-flows and counter-flows).