This review provides an in-depth analysis of current hardware acceleration approaches for image processing and neural network inference, focusing on key operations involved in these applications and the hardware platforms used to deploy them. We examine various solutions, including traditional CPU–GPU systems, custom ASIC designs, and FPGA implementations, while also considering emerging low-power, resource-constrained devices.

The constant increase in the number of space objects and debris orbiting the Earth poses risks to satellites and other spacecraft, both in orbit and during the launching process. Therefore, the monitoring of space hazards to assess risk and prevent collisions has become part of the European Space Policy and requires the establishment of a dedicated Framework for Space Surveillance and Tracking (EU SST) Support. This article presents the CHEIA SST Radar, a new space tracking radar sensor developed and installed in Romania with the purpose of being included in the EU SST sensor network and of contributing to the joint database of space objects orbiting the Earth. The paper describes the processes of design, simulation, and implementation of the hardware and software building blocks that make up the radar system. It emphasizes the particular case of using an already existing system of two large parabolic antennas requiring an innovative retrofitting design to include them as the basis for a new quasi-monostatic radar using LFMCW probing signals. The preliminary design was validated by extensive simulations, and the initial operational testing carried out in December 2021 demonstrated the good performance of the radar in the measuring range and radial speed of LEO space objects.

In this paper, we present an exhaustive description of an extensible e-Health Internet-connected embedded system, which allows the measurement of three biometric parameters: pulse rate, oxygen saturation and temperature, via several wired and wireless sensors residing to the realm of Noncommunicable Diseases (NCDs) and cognitive assessment through Choice Reaction Time (CRT) analysis. The hardware used is based on ATMEGA AVR + MySignals Hardware printed circuit board (Hardware PCB), but with multiple upgrades (including porting from ATMEGA328P to ATMEGA2560). Multiple software improvements were made (by writing high-level device drivers, text-mode and graphic-mode display driver) for increasing functionality, portability, speed, and latency. A top-level embedded application was developed and benchmarked. A custom wireless AT command firmware was developed, based on ESP8266 firmware to allow AP-mode configuration and single-command JavaScript Object Notation (JSON) data-packet pushing towards the cloud platform. All software is available in a git repository, including the measurement results. The proposed eHealth system provides with specific NCDs and cognitive views fostering the potential to exploit correlations between physiological and cognitive data and to generate predictive analysis in the field of eldercare.

This paper describes the work for the radar signal processor, the core of the Cheia space surveillance radar. It presents the basic operation, the requirements and the achieved processing performance together with a description of optimizations both in terms of signal-to-noise ratio and in terms of software processing. The Cheia Radar is financed by the European Space Agency and will be used for the tracking of low Earth orbit objects and for refining the international catalogs of space objects.

Registering UAV-based thermal and visible images is a challenging task due to differences in appearance across spectra and the lack of public benchmarks. To address this issue, we introduce UAV-TIRVis, a dataset consisting of 80 accurately and manually registered UAV-based thermal (640 × 512) and visible (4K) image pairs, captured across diverse environments. We benchmark our dataset using well-known registration methods, including feature-based (ORB, SURF, SIFT, KAZE), correlation-based, and intensity-based methods, as well as a custom, heuristic intensity-based method. We evaluate the performance of these methods using four metrics: RMSE, PSNR, SSIM, and NCC, averaged per scenario and across the entire dataset. The results show that conventional methods often fail to generalize across scenes, yielding <0.6 NCC on average, whereas the heuristic method shows that it is possible to achieve 0.77 SSIM and 0.82 NCC, highlighting the difficulty of cross-spectral UAV alignment and the need for further research to improve optimization in existing registration methods.

This emergence of the heterogeneous computing is based mainly on various forms of parallel accelerators. We present a family of accelerators for embedded computation with a map-reduce architecture based on the partial recursive functions computation model introduced by Stephen Kleene. A three-level virtual prototyping environment is provided to support the development of embedded applications. The first level is written in a Lisp-like functional language. The second is a C-like environment which segregates the intense part of the computation from the complex part. The last one is a low level simulator able to provide support for advanced optimizations. The environment is designed for developing applications by tuning the architecture of a family of many-core machines which provide high performance per Watt and cm2. The energy efficiency of processors backing our architectural approach is in the range of 10 pJ/flop evaluated for the standard cell 28nm technology.