AER Time-domain Vision Sensor Behavioral Modeling And Algorithms Analysis

With wireless video sensor networks, biomedical and autopilot systems and other new applications emerging, acquisition and real-time processing of image information is becoming more and more important, and also machine vision technology based image has been widely used in all kinds of intelligent analysis, identification and control system. With rapid development of the image sensor, spatial resolution and temporal resolution of the image quickly increases, so that the size of image data is also growing exponentially. Collecting and processing based on "frame" not only brings pressure for image transmission and storage, but also proposed to deal with the speed requirements for the real-time machine vision application. In order to overcome the drawbacks of traditional "frame sampling" approach, this paper studies vision sensors based on the Address-Event Representation.The vision sensor based on based on Address-Event Representation theory and neuromorphic engineering theory gets and transports high frame frequency and low data quantity of visual information which is no redundancy. The behavioral emulation of vision sensor is designed. The round arbitration, specific regional first and event concentration first are used by the behavioral emulation. The experimental results and theoretical analysis show that under the same conditions, the AE quantity generated by vision sensor based on AER is less than5%of traditional frame data and the equivalent frame frequency is more than600frame/s. This paper analyzes tracking algorithm and shape recognition algorithms based on the address-event event. Because the traditional tracking algorithm is based on the frame data, and frame data is highly redundant, so the traditional tracking algorithm requires the higher speed processor, and needs large amount of calculation. It cannot achieve high-speed real-time tracing. High-speed object tracking algorithm based on AE data to solve this problem, to achieve the goal of high-speed real-time tracking. Since AE data contains only the outline information of moving object, so the shape recognition algorithm based on AE data can quickly analyze the shape of the object.Vision sensor based on AER is better than conventional image sensor based on frame sampling mode, eliminating redundant data from the source, greatly reduce the complexity and computational of analysis and processing, while pixel asynchronous communication can be greatly improved the real-time processing, and can perform real-time visual information collection, processing and output, suitable for high-speed, high real-time requirements of visual fields.