A Dense People Flow Statistics Method Based on Deep Learning Head Detection

Abstract

A dense crowd counting method based on deep learning head detection: Step 1) Manually collect the surveillance video of the surveillance scene, mark the head data of the scene with the head frame in the surveillance video, and use the deep learning framework to establish a deep residual convolution for head detection Neural network, and train the neural network; Step 2) Input the surveillance video into the trained deep residual convolutional neural network frame by frame in real time to obtain all the head frames in each frame of the surveillance video; Step 3) For the current frame picture, Determine whether each head frame in the picture has been counted. If there is no head frame in the current frame, go to S2; step 4) track and judge the head frame that has not been counted in step 3) frame by frame, and if it is confirmed to be valid The heads are summed up to the total number of heads, otherwise the head frame is discarded.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method for counting dense crowds based on deep learning head detection. Background technique [0002] In recent years, with the increasing demand for security in the society, the deployment of cameras has increased significantly in places with dense passenger flows such as railway stations, subway stations, and airports. In many current monitoring scenarios, the counting of people flow is usually done by manual observation. Due to the large deployment of cameras, the workload of manual observation is also huge, which makes it extremely difficult to make full use of more image data. [0003] Recently, a large number of humanoid detection techniques based on deep learning, including "a deep learning-based passenger flow counting method in a vertical perspective", have been used for passenger flow counting tasks, but they still cannot effectively solve the following probl...

AI Technical Summary

Problems solved by technology

Due to the large deployment of cameras, the workload of manual observation is also huge, making it extremely difficult to make full use of more image data

[0003] Recently, a large number of humanoid detection techniques based on deep learning, including "a deep learning-based passenger flow counting method in a vertical perspective", have been used for passenger flow counting tasks, but they still cannot effectively solve the following problems:

[0004] 1 When the flow of people is dense, there is serious mutual occlusion among pedestrians, and the occlusion of backpacks and other items leads to the problem of missed detection of people

[0005] 2 The background of the actual scene is very complex, and there are a large number of non-human moving objects

Such as elevators going up and down, subways coming in, screens playing various advertisements, etc. These complex moving objects often lead to false detection

[0006] 3 Some surveillance cameras have insufficient resolution. When pedestrians are far away, their images are not clear, which leads to missed detection

Images