Unmanned object detection for image surveillance systems

In this paper we consider an object detection problem for image based surveillance system that works in a predefined area. A real environment is a random spatial and temporal changing background, e.g. water waves, light, wind, rain, fog, etc. The suspicious objects (humans, cars, ships, etc.) are put in an observation area from either an area boundary or its surface. Therefore the object and environment models development and their descriptors determination compose a complicated problem. We have developed principles and methods of intelligent object detection for image surveillance automat, denoted iODA, based on artificial intelligence fuzzy expert systems approach. These methods take into account that a complicated environment data renewal can´t be used. Object detection in complicated random changing environment requires high representative suspicious object descriptors generation, accumulation of some data analysis of partial results and decoding the multistage object detection results for final decision making. The iODA developed method provides an optimization of the needed processing power, processing time and environment complexity. Many experimental results represent the methods functioning. The developed methods are used in some intelligent systems as coastal surveillance system, prison security system and person-animal surveillance system.