| Information age,people pursue more effective data processing methods.The pedestrian detection technology base on the carrier of videos and images has become the focus of the present research for its wide application values.Pedestrian detection in the traditional machine learning field has more restrictive factors and little room for improvement.Depth learning in the image processing field has shown great advantages,can extract more effective information.In recent years,people have tried a variety of different depth-based network of pedestrian detection methods,and achieved good detection results.However,the current pedestrian detection method based on the depth network,most cannot get the detection effect and the detection speed win-win situation.On the one hand,it is necessary to improve the feature extraction capability of the depth model by stacking the network layers,which will lead to the increase of the model parameters.On the other hand,the improvement of the detection speed usually means that the model cannot achieve satisfactory test results.How to achieve an excellent pedestrian detection algorithm both in detection effect and detection speed,become the focus of this paper.Aiming at the above problems,this paper presents a real-time pedestrian detection algorithm based on convolution neural network.Firstly,according to the research results of convolution neural network,a feature extraction model with high network layer number,low parameter and fast running speed is designed by the ideas of simplifying Inception,reducing convolution kernel size and remove full connection layers.Then,based on the above feature extraction model,the design of the pedestrian detection model is completed by using the improved region proposal method.Among them,the application of the regional proposal method is changed from a single feature map to different feature maps.According to the particularity of pedestrians,set more consistent boxes for pedestrian prediction.In order to solve the problem that the smaller pedestrian pixel is easy to missing,this paper puts forward the corresponding improvement method.Finally,the experiment is carried out by using the extended pedestrian test data set ETH + which is proposed in this paper.The experimental results show that the feature extraction model proposed in this paper has similar feature extraction effect and higher computational speed than other classical convolution networks.In this paper,the design of the depth of pedestrian detection methods,compared with other typical algorithms,has a more balance results in detection effect and detection speed.The pedestrian detection algorithm,in ensuring the accuracy of the case,can meet the purpose of real-time detection. |
