• background subtraction
• detection
• infrared
• IRST

Target detection is a crucial assignment in several fields. It aims at managing and protecting the object being supervised for the target could be dangerous, intentionally or not.
In maritime scenarios, target detection can be useful for managing ship routes near the coast both from land and on board of the vessels, for example it can be used from port captaincy in order to avoid ship accidents, or in the open sea, in order to detect hostile behaviours from other boats.
The need of security for military ships is pretty obvious, but it is necessary also for merchant ships, in fact, in the last years, the incidence of pirate attack to merchant ships as increased and the cost of piracy to the world shipping in the global economy is estimated to about £18 billion a year.
Surveillance technology in maritime scenarios are mainly oriented towards active sensors such as radar. Nevertheless, they are very expansive and not always the optimum choice since it makes the user more visible to the distance when, especially for military ships, it is desirable to keep stealth. Furthermore, they produce electromagnetic fields that are harmful for human health and can be affected by lots of electromagnetic compatibility problems with other board instruments.
Infrared cameras are, on the contrary, passive instruments, generally cheaper than radars. Since they do not produce electromagnetic fields, they avoid increasing the visibility of the ship and can be placed almost everywhere.
In order to detect targets in infrared images, several algorithms are used. Most of them are sequence-based and take advantage of the time stationarity of the background to highlight the target by subtraction of consecutive images. These methods are time consuming, because they have to store a buffer of images before producing a result, and provide insufficient results for maritime environment, where the sea clutter is rapidly changing. For this reason, this thesis focus on the frame-based approach. These methods consider the spatial stationarity of the clutter and estimate it by means of frequency domain filters or spatial domain filters in order to delete it.
For this thesis, some linear and nonlinear spatial domain filters have been developed. They require knowing the dimension of the target to be detected. This is not a strong limit for those algorithms because the goal is the first detection of the target, after which it can be followed by a tracker. The first detection desirably happens when the target is far from the camera, near the horizon line, where its capture occupies a window with dimensions of few pixels.
The validation of the proposed algorithms has been done through a dataset of real IR images captured on maritime environment. In this dataset, all the images of each sequence have been analysed as if it was necessary to find the target in each frame and not only the first time. Furthermore, the algorithm has been tested even on nearer target, which occupies a bigger number of pixels.
The large amount of real images taken has permitted not to use simulated images.
In order to evaluate the efficiency throughout all the images it has been necessary to create a small tracking algorithm that automatically recognise the movement of the target.
The results show that the proposed algorithms are useful to detect different sizes and shapes of targets, which means that they are useful not only for the purpose for which they were originally considered, but also for other tasks, such as search and rescue operations, that are nowadays very frequent because of the increasing immigration problem.