Method and device for pedestrian detection based on depth convolutional network

Abstract

The invention discloses a method and a device for pedestrian detection based on a depth convolutional network. The method comprises a convolutional neural network training step and a pedestrian detection step. The convolutional neural network training comprises the following steps: S10, a plurality of groups of sample image data are selected from an image library; S11, one group of sample image data is sent to an input layer of a multilayer neural convolutional network; S12, the output vector of an intermediate layer of the neural convolutional network and the actual output vector of an output layer are calculated to obtain the error of the intermediate layer and the error of the output layer; S13, the weight from an output layer vector element to an intermediate layer output vector element and the weight from the intermediate layer output vector element to the output layer vector element are adjusted; and S14, the total error function value is judged, and pedestrians are detected by use of the trained network. The method and the device of the invention have the advantages of better robustness and high rate of detection accuracy.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a pedestrian detection method and device based on a deep convolutional network. Background technique [0002] With the development of science and technology, intelligent video surveillance technology has gradually become one of the research hotspots of video surveillance technology. The so-called "intelligence" is to fully mine and extract key information in video resources, and use this information to provide users with valuable services. For example, when the monitoring system finds that an item of unknown origin has appeared in a safe area for a long time, or finds suspicious persons wandering in a safe management area or other abnormal behaviors, the system can capture such potentially threatening events in time, And make an intelligent judgment on whether to intervene in the event, so as to effectively suppress a series of problems caused by people as the m...

AI Technical Summary

Problems solved by technology

In addition to the difficulties of clothing changes and posture changes in general human detection, pedestrian detection also has the following difficulties due to its specific application field: the camera is moving, so the method widely used in the field of intelligent monitoring to detect dynamic targets cannot be directly Use; Pedestrian detection is faced with an open environment, and different road conditions, weather and light changes have to be considered, which puts forward high requirements for the robustness of the algorithm; real-time performance is a requirement that the system must meet, which requires the use of Image processing algorithms should not be too complex

Since the traditional algorithms are not adaptable to the above problems, a new method should be proposed to solve the above problems.

Images