Research focus is on the next generation of the image/video processing pipeline, of computer vision applications and the use of deep-learning for deploying smarter digital cameras.

Nowadays we are assisting to create an unprecedented interest in sophisticated imaging/video tools for improving the end-user activities in various real-world applications. This is mainly due to two major factors, the availability of a large variety of acquisition and visualization devices and the convincing performances of deep-learning technology in various imaging and computer vision tasks.  This has opened a large deployment of tools and devices that are extensively used in various applications under different areas, i.e., self-driving, smart cities, monitoring and control of critical infrastructures, entertainment, industry and agriculture, health etc.

However, we are still in an infancy stage where the provided deep-learning approaches are working well under restricted and ideal conditions, while the hardware sensors such as cameras are still limited to the classical optical processes to acquire image/video data. From one end, this is limiting the use of existing deep-learning approaches to real-world applications. On the other hand, sophisticated and expensive cameras sensors can partially overcome the limitations of the optical processes to acquire high quality images/videos data, while cheap camera sensors are by far incapable to acquire robust and reliable images/videos data under complex and different environmental conditions.

What is still missing, is the provision of a seamless integration of robust and reliable level of deep-learning approach, e.g., intelligence, into existing cameras hardware, e.g., high-end/low-end sensors to create what we can call smarter-camera system, as shown in Fig. 1, which provides robust and reliable data acquisition, storage, manipulation, task-driven decision and visualization of the acquire input camera data overcoming the above limitations.

Figure 1 - General concept of the smarter-camera – through the use of different algorithmically solutions, the camera will be capable to overcome any hardware deficiency in the acquisition of the input data, transmit the data on complex network infrastructures with limited data loss, elaborate the information in order to provide robust and reliable decision for a task-driven application, elaborate the input data for reliable visualization as well as will allows the integration with augmented data to help for subsequent managerial decisions.