Neural networks are the state-of-the-art solution to image-processing tasks. Some of these neural networks are relatively simple, but the popular convolutional neural networks (CNNs) can consist of hundreds of layers. Unfortunately, the excellent recognition accuracy of CNNs comes at the cost of very high computational complexity, and one of the current challenges is managing the power, delay and physical size limitations of hardware solutions dedicated to accelerating their inference process. In this paper, we describe the embedding of an eye detection system on a Zynq XCZU4EV UltraScale+ multiprocessor system-on-chip (MPSoC). This eye detector is used in the application framework of a remote iris recognition system, which requires high resolution images captured at high speed as input. Given the high rate of eye regions detected per second, it is also important that the detector only provides as output images eyes that are in focus, discarding all those seriously affected by defocus blur. In this proposal, the network will be trained only with correctly focused eye images to assess whether it can differentiate this pattern from that associated with the out-of-focus eye image. Exploiting the neural network’s advantage of being able to work with multi-channel input, the inputs to the CNN will be the grey level image and a high-pass filtered version, typically used to determine whether the iris is in focus or not. The complete system synthetises other cores and implements CNN using the so-called Deep Learning Processor Unit (DPU), the intellectual property (IP) block released by AMD/Xilinx. Compared to previous hardware designs for implementing FPGA-based CNNs, the DPU IP supports extensive deep learning core functions, and developers can leverage DPUs to conveniently accelerate CNN inference. Experimental validation has been successfully addressed in a real-world scenario working with walking subjects, demonstrating that it is possible to detect only eye images that are in focus. This prototype module includes a CMOS digital image sensor that provides 16 Mpixel images, and outputs a stream of detected eyes as 640 × 480 images. The module correctly discards up to 95% of the eyes present in the input images as not being correctly focused.

Service robotics involves the design of robots that work in a dynamic and very open environment, usually shared with people. In this scenario, it is very difficult for decision-making processes to be completely closed at design time, and it is necessary to define a certain variability that will be closed at runtime. MAPE-K (Monitor–Analyze–Plan–Execute over a shared Knowledge) loops are a very popular scheme to address this real-time self-adaptation. As stated in their own definition, they include monitoring, analysis, planning, and execution modules, which interact through a knowledge model. As the problems to be solved by the robot can be very complex, it may be necessary for several MAPE loops to coexist simultaneously in the robotic software architecture endowed in the robot. The loops will then need to be coordinated, for which they can use the knowledge model, a representation that will include information about the environment and the robot, but also about the actions being executed. This paper describes the use of a graph-based representation, the Deep State Representation (DSR), as the knowledge component of the MAPE-K scheme applied in robotics. The DSR manages perceptions and actions, and allows for inter- and intra-coordination of MAPE-K loops. The graph is updated at runtime, representing symbolic and geometric information. The scheme has been successfully applied in a retail intralogistics scenario, where a pallet truck robot has to manage roll containers for satisfying requests from human pickers working in the warehouse.

Plankton are an important component of life on Earth. Since the 19th century, scientists have attempted to quantify species distributions using many techniques, such as direct counting, sizing, and classification with microscopes. Since then, extraordinary work has been performed regarding the development of plankton imaging systems, producing a massive backlog of images that await classification. Automatic image processing and classification approaches are opening new avenues for avoiding time-consuming manual procedures. While some algorithms have been adapted from many other applications for use with plankton, other exciting techniques have been developed exclusively for this issue. Achieving higher accuracy than that of human taxonomists is not yet possible, but an expeditious analysis is essential for discovering the world beyond plankton. Recent studies have shown the imminent development of real-time, in situ plankton image classification systems, which have only been slowed down by the complex implementations of algorithms on low-power processing hardware. This article compiles the techniques that have been proposed for classifying marine plankton, focusing on automatic methods that utilize image processing, from the beginnings of this field to the present day.

Although the global population is aging, the proportion of potential caregivers is not keeping pace. It is necessary for society to adapt to this demographic change, and new technologies are a powerful resource for achieving this. New tools and devices can help to ease independent living and alleviate the workload of caregivers. Among them, socially assistive robots (SARs), which assist people with social interactions, are an interesting tool for caregivers thanks to their proactivity, autonomy, interaction capabilities, and adaptability. This article describes the different design and implementation phases of a SAR, the CLARA robot, both from a physical and software point of view, from 2016 to 2022. During this period, the design methodology evolved from traditional approaches based on technical feasibility to user-centered co-creative processes. The cognitive architecture of the robot, CORTEX, keeps its core idea of using an inner representation of the world to enable inter-procedural dialogue between perceptual, reactive, and deliberative modules. However, CORTEX also evolved by incorporating components that use non-functional properties to maximize efficiency through adaptability. The robot has been employed in several projects for different uses in hospitals and retirement homes. This paper describes the main outcomes of the functional and user experience evaluations of these experiments.

One of the main challenges faced by iris recognition systems is to be able to work with people in motion, where the sensor is at an increasing distance (more than 1 m) from the person. The ultimate goal is to make the system less and less intrusive and require less cooperation from the person. When this scenario is implemented using a single static sensor, it will be necessary for the sensor to have a wide field of view and for the system to process a large number of frames per second (fps). In such a scenario, many of the captured eye images will not have adequate quality (contrast or resolution). This paper describes the implementation in an MPSoC (multiprocessor system-on-chip) of an eye image detection system that integrates, in the programmable logic (PL) part, a functional block to evaluate the level of defocus blur of the captured images. In this way, the system will be able to discard images that do not have the required focus quality in the subsequent processing steps. The proposals were successfully designed using Vitis High Level Synthesis (VHLS) and integrated into an eye detection framework capable of processing over 57 fps working with a 16 Mpixel sensor. Using, for validation, an extended version of the CASIA-Iris-distance V4 database, the experimental evaluation shows that the proposed framework is able to successfully discard unfocused eye images. But what is more relevant is that, in a real implementation, this proposal allows discarding up to 97% of out-of-focus eye images, which will not have to be processed by the segmentation and normalised iris pattern extraction blocks.

There exist image processing applications, such as tracking or pattern recognition, that are not necessarily precise enough to maintain the same resolution across the whole image sensor. In fact, they must only keep it as high as possible in a relatively small region, but covering a wide field of view. This is the aim of foveal vision systems. Briefly, they propose to sense a large field of view at a spatially-variant resolution: one relatively small region, the fovea, is mapped at a high resolution, while the rest of the image is captured at a lower resolution. In these systems, this fovea must be moved, from one region of interest to another one, to scan a visual scene. It is interesting that the part of the scene that is covered by the fovea should not be merely spatial, but closely related to perceptual objects. Segmentation and attention are then intimately tied together: while the segmentation process is responsible for extracting perceptively-coherent entities from the scene (proto-objects), attention can guide segmentation. From this loop, the concept of foveal attention arises. This work proposes a hardware system for mapping a uniformly-sampled sensor to a space-variant one. Furthermore, this mapping is tied with a software-based, foveal attention mechanism that takes as input the stream of generated foveal images. The whole hardware/software architecture has been designed to be embedded within an all programmable system on chip (AP SoC). Our results show the flexibility of the data port for exchanging information between the mapping and attention parts of the architecture and the good performance rates of the mapping procedure. Experimental evaluation also demonstrates that the segmentation method and the attention model provide results comparable to other more computationally-expensive algorithms.