Dynamic Data Driven Applications Systems (DDDAS) is a transformative framework for incorporating evolving data into a dynamic system to adapt to operational conditions, and recursively to steer its measurement components that generate such data. Signal processing techniques are applied to big data analysis to spawn developments in designs, algorithms, architectures, and applications. DDDAS is based on four interactive concepts: applications modeling, mathematical and statistical processing, measurement systems, and systems software design, as shown in Fig. 1. Rooted in control theory, many of the prominent developments include methods of signal processing to determine the relevant data for model updates. These model updates arise from statistical data analysis. New insights have been developed for measurement systems with unobservable data, such as new architectures to emulate data collections for missing data. To complement the modeling and statistical analysis, software methods are paramount for real-time applications. With recent advances from Very Large Scale Integration (VLSI) and high performance computing (HPC), implementations of DDDAS concepts are realizable. The DDDAS framework has spawned numerous applications such as environment analysis (e.g. weather); robotic systems (e.g., unmanned aerial vehicle, unmanned ground vehicle (UAV/UGV) coordination); image processing (e.g., target tracking), and embedded computing (e.g., hardware/software designs). This special issue brings together DDDAS examples for the readers. More information can be found at www.1dddas.org.

Figure 1
figure 1

Dynamic data-driven applications systems (DDDAS) framework.

The special issue received numerous submissions and two were selected based on the implementation of the entire DDDAS framework presented in Fig. 1.

The first was S. Sarkar, et al., “Deep Learning for Automated Occlusion Edge Detection in RGB-D Frames” (10.1007/s11265-016-1209-3). The authors addressed the challenge of real-time video tracking when objects are occluded. Using Deep Convolutional Neural Networks (CNN) algorithms from computer vision, they demonstrate robust occlusion edge determination via DDDAS processing of heterogeneous measurements from visual cameras, depth modeling, and motion models. The system level performance application was executed in experiments using parallel computing software with Graphical Processor Units (GPUs) from the Compute Unified Device Architecture (CUDA) reducing the false alarms from occlusion detection.

The second was R. Wu, et al., “A Container-based Elastic Cloud Architecture for Pseudo Real-time Exploitation of Wide Area Motion Imagery (WAMI) Stream” (10.1007/s11265-016-1206-6)”. The authors addressed the difficulty of using large scale data as demonstrated with WAMI. The paper highlights software implementation solutions by using a cloud-based approach to decompose the imagery and enhance multi-target tracking. Their DDDAS approach included the Pseudo Real-time Exploitation of Sub-Area (PRESA) framework which divides the data into spatial regions for parallel processing of video tracking algorithms. Key to the approach was the stitching of tracks from the spatial regions. The results demonstrate that the virtual-machine container approached improved the processing frame rate, object detection, and tracking accuracy, as compared to approaches using the Hadoop method.

The guest editors believe these papers demonstrate solutions using the DDDAS framework to advance the field of signal processing systems and hope they stimulate further development of DDDAS applications. The editors express their appreciation to the authors and reviewers for contributing to this special issue.