Pedestrian counting method based on deep neural network

Abstract

The invention discloses a pedestrian counting method based on a deep neural network. The pedestrian counting method comprises the following steps: acquiring a video frame image by using a network camera; performing target detection by using a deep neural network to obtain a pedestrian detection frame; tracking the pedestrian frame by using a tracking algorithm to obtain motion tracks of differentpedestrians; and analyzing the motion track of the pedestrian by utilizing a counting algorithm so as to judge the walking direction of the pedestrian and when the counting is increased. According tothe invention, monitoring and counting can be carried out on pedestrian crowded scenes and pedestrian sparse scenes, and the practical application value is very high, and the calculated amount is small, and the false identification rate is low, and as only a single camera is used, the installation is convenient, and the influence on the monitoring environment is very small.

Description

technical field [0001] The invention relates to the field of deep computer vision, in particular to a pedestrian counting method based on a deep neural network. Background technique [0002] In the field of intelligent video surveillance technology, the research on pedestrian counting has always been the core of attention of those skilled in the art. This is because, by counting the pedestrians in the scene, when an emergency occurs, it can effectively disperse and evacuate the flow of people according to the distribution of the number of people in the scene, and minimize the harm caused by the emergency. [0003] At present, there are mainly four methods for pedestrian counting: [0004] Based on pressure sensor technology: Usually the pressure sensor will be installed in the heavy area where pedestrians pass. When pedestrians walk through the sensor detection area, different electrical signals will be generated due to different weights. Passenger flow statistics can be re...

AI Technical Summary

Problems solved by technology

[0004] Based on pressure sensor technology: Usually the pressure sensor will be installed in the heavy area where pedestrians pass. When pedestrians walk through the sensor detection area, different electrical signals will be generated due to different weights. Passenger flow statistics can be realized through changes in electrical signals, but Because the equipment is often trampled, the long-term passage of high-density large passenger flow will seriously affect the life of the equipment and reduce the detection accuracy

[0005] Pedestrian detection method based on infrared detection device: When pedestrians pass through the infrared detection device, the receiving device cannot receive it, and count at this time. At present, this method cannot accurately count parallel pedestrians, and is not suitable for high-density scenes

[0006] Based on radio wave pedestrian detection technology: radio wave detection of pedestrians is mainly to calculate the attenuation of fixed-frequency monitoring radio waves in the channel where people pass through, and calculate the attenuation through professional equipment to estimate the number of passengers in the channel. This method can only Calculate a quantity range, and the detection scene is relatively fixed, it is difficult to monitor the station as a whole, and the purchase of professional equipment is expensive

[0007] Pedestrian counting method based on video image analysis: it can be mainly divided into traditional image processing method and deep learning processing method. Traditional image processing uses artificially designed features suitable for detecting people, and judges whether pedestrians exist by detecting whether these features are present in the picture. , and then use the detected "person" features to track the target to achieve the purpose of counting. This method is applicable in the scene of sparse pedestrians. Once the crowd is dense and there is occlusion between pedestrians, the detection accuracy will drop

Images