Fly's Vision Neuron Based Objective Motion Detection And Tracking

Fly's vision systems are extremely sensitive to moving objects, being capable of providing us a rich source of biological information in studying the problem of objective tracking and detection on position and velocity. It also give us many biologically theoretical foundations for exploring artificial fly visual neural networks. Based on such considerations, this thesis studies several improved artificial fly visual neural network models for the problem of objective detection and tracking under static or dynamic environments, depending on image processing. It also investigates several algorithms of directional detection and objective tracking. All these models and algorithms are examined by means of video image sequences under different actual situations.The research achievements acquired are summarized below.1. For the problem of motion-directional detection of a single object involved in an image sequence under static environments, an object extraction approach is designed, relying upon some methods of background subtraction and morphological processing. Thereafter, a motiondirectional detection model is developed based on the principle of motion detection associated to the inner structures of compound eyes.2. For the problem of tracking and detection on position and motion direction for multiple objects under static environments, a reported object extraction approach is adopted to eliminate background noises by virtue of inter-frame difference and morphological processing. In view of imperfect object extraction, a multiple objective recognizing method is designed to decide the numbers of objects and their motion locations in the picture. Related to the inner structure mechanism of fly's visual nerves, an improved artificial fly visual neural network model is developed to detect the locations of the objects and to execute directional detection.3. For the problem of motion-directional detection of a single object under dynamic environments, an inter-frame difference method is selected to detect brightness variation between two adjacent frames. An improved artificial fly visual neural network model capable of strengthening the robustness of objective detection is developed to output the moving direction of the object included in a video sequence with background interference, after modifying the design of Medulla layer of a reported visual neural network model.