Real-time Wearable Computer Vision System for Improved Museum Experience

Computer vision system for automated cell pressure probe operation

Real-time underwater sorting of edible fish species

Real-time image processing and computer vision systems are now in the mainstream of technologies enabling applications for cyber-physical systems, Internet of Things, augmented reality, and Industry 4.0. These applications bring the need for Smart Cameras for local real-time processing of images and videos. However, the massive amount of data to be processed within short deadlines cannot be handled by most commercial cameras. In this work, we show the design and implementation of a manycore vision processor architecture to be used in Smart Cameras. With massive parallelism exploration and application-specific characteristics, our architecture is composed of distributed processing elements and memories connected through a Network-on-Chip. The architecture was implemented as an FPGA overlay, focusing on optimized hardware utilization. The parameterized architecture was characterized by its hardware occupation, maximum operating frequency, and processing frame rate. Different configurations ranging from one to eighty-one processing elements were implemented and compared to several works from the literature. Using a System-on-Chip composed of an FPGA integrated into a general-purpose processor, we showcase the flexibility and efficiency of the hardware/software architecture. The results show that the proposed architecture successfully allies programmability and performance, being a suitable alternative for future Smart Cameras.

The paper is devoted to the problem of formalization of software-hardware solutions in designing real-time computer vision systems. The main attention is paid to the methods of implementation of low-level operations that find features (simple elements) in the image input to the system. Algorithmic types of detectors of simple elements in images are analyzed from the point of view of hardware organization in computer vision systems. In this connection, the necessary performance and memory resources are estimated. The capabilities of parallel and pipeline execution of detector algorithms are investigated. A method of using a field programmable gate array and a digital signal processor in solving the problem of image processing in real-time computer vision systems is considered in detail.

Continuous real-time monitoring and neural network modeling of apple slices color changes during hot air drying

One of the fundamental problems in computer vision is the background-foreground segmentation and most of the strategies have severe drawbacks when working with natural images, where there are extreme conditions such as illumination changes combined with sudden background differences or other noise; moreover if the system has to face real-time restrictions. In this case authors focus on a variation of the U-net architecture to obtain the segmentation of the objects (fishes) in every single frame. The U-net has some interesting properties to explore in the case of image segmentation, such as multi-scale parameter combination. The reported preliminary results, working in a context of a multi-target fish tracking are promising, and envisions an approach that could provide a real-time response to long-lasting experiments using HQ video for multi-target tracking in real-time Computer vision systems.

TOUCH is a real-time computer vision system that has been developed in order to aid the operation of autonomous underwater vehicles (AUVs) and remotely operated vehicles (ROVs). The algorithm is based on a modified 4-connectivity approach that scans the incoming frames for a target hue range and based upon the largest connected blobs of hue, tracks an object and provides the two dimensional Cartesian coordinates of the desired object. Since few algorithms exist for tracking and recognizing objects in underwater environments, this algorithm provides a way for AUVs and ROVs to acquire, track or recognize objects by using a relatively fast and inexpensive vision system. This paper presents the details on TOUCH, its testing, some areas that warrant improvement and possible course of action for the future.

Local scale information extracted from visual data in a bottom-up manner constitutes an important cue for a large number of visual tasks. This article presents a framework for how the computation of such scale descriptors can be performed in real time on a standard computer. The proposed scale selection framework is expressed within a novel type of multi-scale representation, referred to as hybrid multi-scale representation, which aims at integrating and providing variable trade-offs between the relative advantages of pyramids and scale-space representation, in terms of computational efficiency and computational accuracy. Starting from binomial scale-space kernels of different widths, we describe a family pyramid representations, in which the regular pyramid concept and the regular scale-space representation constitute limiting cases. In particular, the steepness of the pyramid as well as the sampling density in the scale direction can be varied. It is shown how the definition of γ-normalized derivative operators underlying the automatic scale selection mechanism can be transferred from a regular scale-space to a hybrid pyramid, and two alternative definitions are studied in detail, referred to as variance normalization and lp-normalization. The computational accuracy of these two schemes is evaluated, and it is shown how the choice of sub-sampling rate provides a trade-off between the computational efficiency and the accuracy of the scale descriptors. Experimental evaluations are presented for both synthetic and real data. In a simplified form, this scale selection mechanism has been running for two years, in a real-time computer vision system.

In this paper, we present an autonomous helicopter with vision based navigation called South China University of Technology unmanned aerial vehicle (SCUAV). A GPS/INS system has been designed and implemented for getting stable navigation information. A Kalman filtering has been used in this system for data fusion. A real-time computer vision system is presented in this paper as the complement of the GPS/INS system. The vision algorithm is designed and implemented in this paper, which is integrated with algorithms for tracking a known landmark and estimating the helicopter positions. A method of image processing is designed for tracking and recognizing known land marks. At the end of the paper, we will present the experiment results to demonstrate our efficacious algorithm.

During the development of a real time computer vision system specialized in tire canvas transversal inspection, it became clear that its most critical primitive is the unsupervised identification of canvas parameters namely colors and threads size. As the real time imposition disables the use of established computer vision primitives, a new approach based in independent component analysis (ICA) combined with 1-D discrete Fourier analysis is introduced.

The advances in the development and application of computer hardware, software, and electronic technologies and control systems provides strong support to fast, consistent measurements, data collection, and information analysis for solving problems in the drying industry. One of the most usable and prominent methods and technology used in the drying industry is computer vision technology. Modern real-time computer vision systems (CVSs) can be the best alternative for ill-defined processes monitoring, automating, and controlling due to its rapidness, cheapness, nondestructiveness, sensitiveness, and preciseness. The chapter reviews the current status of CVS use in fruit drying technology, with particular reference to apple and kiwifruit drying processes, and highlights the importance of applying intelligent control systems to solve food quality and energy hindrances. For extraction of quality parameters, the CVS monitoring process must be done in five steps: image acquisition, image processing, feature extraction, pattern recognition, and finally decision making.

Deep learning has revolutionized the performance of many computer vision systems in recent years. In particular deep convolutional neural networks have demonstrated ground-breaking performance in object classification from imagery. However, these techniques typically require sizeable volumes of training data in order to derive the large number of parameters that these types of approaches must learn. In many situations sufficient volumes of imagery for the object types of interest are unavailable. One solution to this problem is to use an initial training set which has properties similar to those of the objects of interest, but is available with a large number of labelled examples. These can be used for initial network training and the resulting partially-learned solution may subsequently be tuned using a smaller sample of the actual target objects. This type of approach, transfer learning, has shown considerable success in conventional imaging domains. Unfortunately, for Synthetic Aperture Radar imaging sensors, large volumes of labelled training samples of any type are hard to come by. The challenge is exacerbated when variations in imaging geometry and sensor configuration are taken into account. This paper examines the use of simulated SAR imagery in pre-training a deep neural network. The simulated imagery is generated using a straightforward process which has the capability to generate sufficient volumes of training exemplars in a modest amount of time. The samples generated are used to train a deep neural network which is then retrained using a comparatively small volume of MSTAR SAR imagery. The value of such a pre-training process is assessed using techniques to explain model performance by visualization. The assessment highlights some interesting aspects of the MSTAR SAR image set with regard to bias.

There many applications throughout industry and government requiring real-time computing. Any application that monitors and/or controls a process would fit into this category. Some examples are: Nuclear power plants, Steel mills, Space program, etc. General Atomics uses eight real-time computer systems for control and high speed data acquisition required to run the nuclear fusion experiments. Real-Time computing can be defined as the ability to respond to asynchronous external events in a predictable (preferably fast) time frame. Real-Time computer systems are similar to other computers in many ways and may by used for general computing requirements such as Time-Sharing. However special hardware, operating systems and software had to be developed to meet the requirement for real-time computing. Traditionally, real-time computing has been a realm of proprietary operating systems with real-time applications written in FORTRAN and assembly language. In the past, these systems adequately served the needs of the real-time world. Many of these systems that were developed 15 years ago are still being used today. However the real-time world is now changing, demanding new systems to be developed. This paper gives a description of general real-time computer systems and how they differ from other systems. However, the main purpose of this paper is to give a detailed technical description of the hardware and operating systems of an existing proprietary system and a real-time UNIX system. The two real-time computer systems described in detail are Modcomp Classic III/95 with the MAX IV operating system and Modcomp TRI-D 9750 with the REAL/IX.2 operating system.

The concept of real time has different meanings in the systems and theory communities. Thus, the existing formal real-time models do not capture all the practically relevant aspects of such computations. This article proposes a new definition that, we believe, allows a unified treatment of all practically meaningful variants of realtime computations. We use the developed formalism to mode two important features of real-time algorithms, namely the presence of deadlines and the real-time arrival of input data. We also emphasize the expressive power of our model by using it to formalize aspects from the areas of real-time database systems and ad hoc networks. We offer formulations of the recognition problem for real-time database systems and of the routing problem in ad hoc networks. Finally, we suggest a variant of our formalism that is suited for modelling parallel distributed real-time algorithms. We believe that the proposed formalism is a first step towards a unified and realistic complexity theory for real-time parallel computations.

The importance of real-time computing is constantly growing. It is required in a large number of systems, including but not limited to aerospace, defense, industrial automation, nuclear engineering, and decision support. The author defines a real-time computing system as a system capable of receiving inputs from an external process, performing the required data processing, and outputting the correct response back to the process fast enough to meet process requirements. In addition, a real-time system must be able to respond to external interrupts generated by I/O devices and other systems. Therefore, a real-time system can be viewed as a three-stage pipeline: data acquisition from sensors or other input devices, data processing, and output to actuators and/or displays. A high-performance real-time system can then be characterized by three key features: superior CPU computational speed; high I/O throughput; and efficient interrupt handling capability. >