The Research On Visual Features Binding Method Based On PCNN

With the continuous development of cognitive psychology, biological neurology and other disciplines, it has been found that there is a great difference between the existing visual computing theory and mechanism of biological vision, Visual computing theory is difficult to show a lot of biological visual features accurately. Therefore it is important to study the mechanism of biological vision, explore mathematical laws of its structure and working mechanism, and find its Incentive status under the task state. For the understanding of the law of biological vision and realizing the machine vision with biological vision, it has important scientific significance and practical value.This paper introduces Pulse Coupled Neural Network(PCNN) model, and carries out visual features based on vector bundles PCNN research on the basis of simulating the visual perception of humans and other higher animal; In addition, this paper introduces image recognition based on convolutional neural network model, combines with image matching technology, fuses depth image information, then carries out the application research of visual feature binding in the obstacle avoidance robot, the main contents are as follows:1) Constructed visual feature binding model based on vector PCNN. PCNN model is the third generation artificial network model, which simulates pulses generated by biological stimulation synchronously, and constructs network with physiological characteristics. The visual feature binding model based on vector PCNN is constructed on the basis of PCNN, and it can better simulate biological feature to achieve image recognition and binding in color and shape.2) Studied the application of visual feature binding in the robot obstacle avoidance. Use visual feature binding and reinforcement learning theory, based on PCNN, Convolutional Neural Networks(CNNs) and image matching technology, integrate with the depth of the image information, learn the prior knowledge of the complexity environment, based on feature binding mechanism, and the results are fed back to obstacle avoidance of the robot to make the smart, fast, efficient and accurate decisions.