Search

To solve the problem of the sensorless control method of a permanent magnet synchronous motor, based on the study of a mathematical model for a permanent magnet synchronous motor and model adaptation theory, a reference model equation and adjustable model equation are derived according to the stator current equation. The correctness of the selected linear compensator matrix is strictly proved. Then, Popov’s super-stability theory is used to derive the speed adaptive law and prove its asymptotic stability. Based on the voltage closed-loop feedback MTPA weak magnetic control strategy, a simulation model of a MRAS control system based on stator current is built and combined with the principle of MRAS. Aiming to investigate the problem that the PI adaptive law in the traditional MRAS algorithm is not robust, super-twisting sliding mode control is introduced to replace the PI adaptive law. The observer based on STSM−MRAS is designed. The simulation model of the MRAS control system based on the super-twisting sliding mode is established. Under certain working conditions, the STSM−MRAS algorithm and the traditional MRAS algorithm are simulated and compared. The results show that the STSM−MRAS algorithm can improve the steady-state performance and robustness of a sensorless control system.

Generating high-quality and diverse human images is an important yet challenging task in vision and graphics. However, existing generative models often fall short under the high diversity of clothing shapes and textures. Furthermore, the generation process is even desired to be intuitively controllable for layman users. In this work, we present a text-driven controllable framework, Text2Human, for a high-quality and diverse human generation. We synthesize full-body human images starting from a given human pose with two dedicated steps. 1) With some texts describing the shapes of clothes, the given human pose is first translated to a human parsing map. 2) The final human image is then generated by providing the system with more attributes about the textures of clothes. Specifically, to model the diversity of clothing textures, we build a hierarchical texture-aware codebook that stores multi-scale neural representations for each type of texture. The codebook at the coarse level includes the structural representations of textures, while the codebook at the fine level focuses on the details of textures. To make use of the learned hierarchical codebook to synthesize desired images, a diffusion-based transformer sampler with mixture of experts is firstly employed to sample indices from the coarsest level of the codebook, which then is used to predict the indices of the codebook at finer levels. The predicted indices at different levels are translated to human images by the decoder learned accompanied with hierarchical codebooks. The use of mixture-of-experts allows for the generated image conditioned on the fine-grained text input. The prediction for finer level indices refines the quality of clothing textures. Extensive quantitative and qualitative evaluations demonstrate that our proposed framework can generate more diverse and realistic human images compared to state-of-the-art methods., Comment: SIGGRAPH 2022; Project Page: https://yumingj.github.io/projects/Text2Human.html, Codes available at https://github.com/yumingj/Text2Human

The egocentric video provides a unique view of event participants to show their attention, vision, and interaction with objects. In this paper, we introduce Ego-Deliver, a new large-scale egocentric video benchmark recorded by takeaway riders about their daily work. To the best of our knowledge, Ego-Deliver presents the first attempt in understanding activities from the takeaway delivery process while being one of the largest egocentric video action datasets to date. Our dataset provides a total of 5,360 videos with more than 139,000 multi-track annotations and 45 different attributes, which we believe is pivotal to future research in this area. We introduce the FS-Net architecture, a new anchor-free action detection approach handling extreme variations of action durations. We partition videos into fragments and build dynamic graphs over fragments, where multi-fragment context information is aggregated to boost fragment classification. A splicing and scoring module is applied to obtain final action proposals. Our experimental evaluation confirms that the proposed framework outperforms existing approaches on the proposed Ego-Deliver benchmark and is competitive on other popular benchmarks. In our current version, Ego-Deliver is used to make a comprehensive comparison between algorithms for activity detection. We also show its application to action recognition with promising results. The dataset, toolkits and baseline results will be made available at: https://egodeliver.github.io/EgoDeliver_Dataset/

Generating plausible hair image given limited guidance, such as sparse sketches or low-resolution image, has been made possible with the rise of Generative Adversarial Networks (GANs). Traditional image-to-image translation networks can generate recognizable results, but finer textures are usually lost and blur artifacts commonly exist. In this paper, we propose a two-phase generative model for high-quality hair image synthesis. The two-phase pipeline first generates a coarse image by an existing image translation model, then applies a re-generating network with self-enhancing capability to the coarse image. The self-enhancing capability is achieved by a proposed structure extraction layer, which extracts the texture and orientation map from a hair image. Extensive experiments on two tasks, Sketch2Hair and Hair Super-Resolution, demonstrate that our approach is able to synthesize plausible hair image with finer details, and outperforms the state-of-the-art.

A lane-level, high-definition (HD) digital map is needed for autonomous cars to provide safety and security to the passengers. However, it continues to prove very difficult to produce error-free maps. To avoid the deactivation of autonomous driving (AD) mode caused by map errors, ensuring map data quality is a crucial task. We propose an ontology-based workflow for HD map data quality assurance, including semantic enrichment, violation detection, and violation handling. Evaluations show that our approach can successfully check the quality of map data and suggests that violation handling is even feasible on-the-fly in the car (on-board), avoiding the autonomous driving mode’s deactivation.

Autonomous Driving (AD) systems use digital maps as a virtual sensor to perceive the environment around the car. As the field of digital maps continues to evolve, existing solutions face new challenges such as integration ability for new map formats (e.g., High Definition maps), supporting onboard and offboard deployment and providing a generic interface to access the road environmental knowledge. In this paper, we propose a knowledge architecture layer for environmental modeling and distinguish between low-level ontologies based on various map data formats and a high-level ontology for representing a generic road environment. The adequacy of the modeling is validated over two use cases: lane change notification and logical inconsistency detection. The performance is measured using real map data and it shows encouraging results for future development within onboard and offboard systems.

Video emotion recognition is a challenging task due to complex scenes and various forms of emotion expression. Most existing works focus on fusing multiple features over the whole video clips. According to our observations, given a long video clip, the emotion is usually presented by only several actions/objects in a few short snippets, and the meaningful cues are buried in the noisy background. When human judging the emotion in videos, we first find the informative clips and then closely look for emotional cues in the frames. In this paper, we propose Dual Focus Attention Network to mimic this process. First, three kinds of features including action, object, and scene are extracted from videos. Second, Two attention modules are used to focus on the visual features of the videos from temporal and spatial dimensions respectively. With our dual focus attention network, we can effectively discover the most emotional frames along the time dimension and the most emotional visual cues in each frame. Our experiments conducted on two widely used datasets Ekman and VideoEmotion show that our proposed approach outperforms the existing approaches.

Despite its high potential, the building's sector lags behind in reducing its energy demand. Tremendous savings can be achieved by deploying building management services during operation, however, the manual deployment of these services needs to be undertaken by experts and it is a tedious, time and cost consuming task. It requires detailed expert knowledge to match the diverse requirements of services with the present constellation of envelope, equipment and automation system in a target building. To enable the widespread deployment of these services, this knowledge-intensive task needs to be automated. Knowledge-based methods solve this task, however, their widespread adoption is hampered and solutions proposed in the past do not stick to basic principles of state of the art knowledge engineering methods. To fill this gap we present a novel methodological approach for the design of knowledge-based systems for the automated deployment of building management services. The approach covers the essential steps and best practices: (1) representation of terminological knowledge of a building and its systems based on well-established knowledge engineering methods; (2) representation and capturing of assertional knowledge on a real building portfolio based on open standards; and (3) use of the acquired knowledge for the automated deployment of building management services to increase the energy efficiency of buildings during operation. We validate the methodological approach by deploying it in a real-world large-scale European pilot on a diverse portfolio of buildings and a novel set of building management services. In addition, a novel ontology, which reuses and extends existing ontologies is presented. The authors would like to gratefully acknowledge the generous funding provided by the European Union’s Horizon 2020 research and innovation programme through the MOEEBIUS project under grant agreement No. 680517.

Recent studies have shown that DNNs are vulnerable to adversarial examples which are manipulated instances targeting to mislead DNNs to make incorrect predictions. Currently, most such adversarial examples try to guarantee “subtle perturbation” by limiting the \(L_p\) norm of the perturbation. In this paper, we propose SemanticAdv to generate a new type of semantically realistic adversarial examples via attribute-conditioned image editing. Compared to existing methods, our SemanticAdv enables fine-grained analysis and evaluation of DNNs with input variations in the attribute space. We conduct comprehensive experiments to show that our adversarial examples not only exhibit semantically meaningful appearances but also achieve high targeted attack success rates under both whitebox and blackbox settings. Moreover, we show that the existing pixel-based and attribute-based defense methods fail to defend against SemanticAdv. We demonstrate the applicability of SemanticAdv on both face recognition and general street-view images to show its generalization. We believe that our work can shed light on further understanding about vulnerabilities of DNNs as well as novel defense approaches. Our implementation is available at https://github.com/AI-secure/SemanticAdv .

Digital map data is an important source of information for the perception of the environment around cars for advanced driver assistance functions. These functions use map data to acquire information about the road infrastructure beyond the visual horizon of the driver. Embedded software components in today's cars typically use code-based processing of the map data to offer this support to advanced driver assistance functions, but the complexity of automotive systems continues to grow towards the realization of autonomous driving. To facilitate the representation and extraction of knowledge, we explore the feasibility of using ontologies for modelling and processing the map data in cars. We describe the challenges of adequately modelling the knowledge and present a proof of concept implementation that is used in a PC-based simulation to evaluate the knowledge extraction capabilities of this approach considering the requirements of representative advanced driver assistance functions.

In severely cold regions, it is very difficult to prevent the formation of ice on solid surfaces, even on superhydrophobic (SH) surfaces. Compared with anti-icing surfaces, icephobic surfaces are more useful in practice. In this work, a traditional epoxy coating was modified by using maleic anhydride (MAH) as a crosslinking agent, in addition to filling a small amount of epoxy resin grafted with fluorine containing chains (FEP). The wettability and the mechanical properties of the modified coating, as well as the adhesive strength of ice on the coating surface and the stability of the icephobic properties of the coating, were systematically studied. The results showed that the modified coating surface became more hydrophobic and the force required to remove ice from the surface was less than that needed for removing on the unmodified coating surface. Furthermore, the modified coating has excellent mechanical properties. After sanding with abrasive paper, the unmodified coating surface is more hydrophilic, whereas the modified coating surface retains its hydrophobicity. X-ray photoelectron spectroscopy (XPS) indicates that compared with crosslinking by traditional polyamides, crosslinking using MAH greatly reduces the number of polar groups on the coating surface. The FEP introduction further reduces the free energy of the coating surface. As a result, the modified epoxy coating exhibits excellent icephobic performance. This study will be helpful for the design of icephobic epoxy coatings for use in practical applications. For example, it may make the de-icing on the underside of trains running in cold regions becomes more easily.

The Indoor Environmental Quality (IEQ) in a building affects occupants’ well-being and productivity. Traditionally, models are developed to predict IEQ satisfaction from physical measurements. These approaches work fine in a laboratory environment but tend to fail in real-world applications. Recent work focuses on collecting direct human feedback on IEQ. However, existing approaches either lack the ability to capture the multiple dimensions of IEQ or the integration with other domain knowledge, e.g. from building information modeling or building automation systems. To tackle this problem, we have developed a novel approach based on Semantic Web Technologies (SWT) which enable interoperability and reasoning. In this paper, the HBC (Human Comfort in Building) ontology is presented, which formally specifies the domain of IEQ in multiple dimensions and relates it to adjacent domains. An online survey is designed in order to specify ontology requirements and collect human feedbacks for evaluating the ontology. We evaluate the use of the HBC ontology in two use cases in an o ce building: recommendation of spaces based on IEQ factors and recommendation of settings for technical equipment from collected feedbacks.

Locating actions in long untrimmed videos has been a challenging problem in video content analysis. The performances of existing action localization approaches remain unsatisfactory in precisely determining the beginning and the end of an action. Imitating the human perception procedure with observations and refinements, we propose a novel three-phase action localization framework. Our framework is embedded with an Actionness Network to generate initial proposals through frame-wise similarity grouping, and then a Refinement Network to conduct boundary adjustment on these proposals. Finally, the refined proposals are sent to a Localization Network for further fine-grained location regression. The whole process can be deemed as multi-stage refinement using a novel non-local pyramid feature under various temporal granularities. We evaluate our framework on THUMOS14 benchmark and obtain a significant improvement over the state-of-the-arts approaches. Specifically, the performance gain is remarkable under precise localization with high IoU thresholds. Our proposed framework achieves mAP@IoU=0.5 of 34.2%.

In this paper, we report our recent work on developing a testbed for the unmanned aerial vehicle (UAV)-assisted unmanned ground vehicle (UGV) system. In sharp contrast to most UGV models in the literature, our system does not require Global Positioning System (GPS) by taking full advantage of the latest technological breakthroughs in UAV. By exploiting the wide-angle camera mounted on the UAV, the system can collect detailed ground information including UGV and destination positions before the UAV sends steering and speed update instructions to the UGV via wireless communications. To optimize the route design from UGV to its destination, we develop various route optimization algorithms by taking into account the position of the anticipated next destination using a probabilistic model. Experimental and computer simulation results confirm the validity of our testbed design and route optimization algorithms.

Videos are dominant on the Internet. Current systems to process large-scale videos are suboptimal due to the following reasons: (1) machine learning modules such as feature extractors and classifiers generate huge intermediate data and place heavy burden to the storage and network, and (2) task scheduling is explicit; manually configuring the machine learning modules on the cluster is tedious and inefficient. In this work, we propose Elastic Streaming Sequential data Processing system (ESSP) that supports automatic task scheduling; multiple machine learning components are automatically parallelized. Further, our system prevents extensive disc I/O by applying the in-memory dataflow scheme. Evaluation on real-world video classification datasets shows many-fold improvements.