Specific Object Detection And Recognition Based On Neural Network

Neutral network has developed rapidly and made great progress in recent years.It inspires researchers from different fields to learn more about neural network.There have been many successful cases of using neural network in the field of target detection and recognition.In this paper,neural network was used to recognize verified code and detect pedestrians respectively.Desired results were achieved for both of these two applications.The process of verification code recognition is mainly divided into three steps: image preprocessing,character segmentation and character recognition,where character segmentation is the most difficult one.In this paper,the projection method was improved to overcome the shortcoming that a single character is easily to be divided into multiple parts.A method to determine the boundary of the character by time delay is adopted to improve the projection method.A good correction result was obtained through this method.The results of BP neural network and deep convolution neural network were compared and analyzed under the condition of small sample.Finally,the more cost-effective BP neural network was selected to recognize verification code.In this paper,the process of pedestrian detection is divided into two parts: pedestrian proposal generation and pedestrian recognition.Based on a scene that has stable background distribution,a novel method of generating pedestrian detection proposals quickly is proposed by using online Gaussian model(this method is referred to as OL_GMPG).High recall rate can be achieved by generating few pedestrian detection proposals through Gaussian model fitting method.Both the positions that people appear most frequently and the scale information of corresponding targets can be obtained through the learning and updating processes of Gaussian model.And this information has an effect on subsequent process of pedestrian recognition or pedestrian tracking positively and significantly.In this paper,the deep convolution neural network is used to recognize pedestrian targets.Both the size of the convolution kernel and the number of layers of the network has significant effect on the test results and training efficiency.Through the comparative analysis,a CNN model with depth of 7 and convolution kernel size of 9 * 9 was used for training and recognition,and finally got a good test result.