Population statistics system based on video surveillance image processing

Abstract

Embodiments of the present application disclose a population statistics system based on video surveillance image processing, and the system comprises a memory, a processor and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the following steps are completed: training three convolutional neural networks with face images of corresponding sizes respectively; three trained convolutional neural networks are connected in parallel, and the outputs of the three parallel convolutional neural networks are all connected tothe same output layer. Three scaled images are input into each convolutional neural network for detection. Three parallel convolutional neural networks map the three images with face detection boxes to the original image through output layer, so that each face includes several detection boxes. The three parallel convolutional neural networks map the three images with face detection boxes to the original image. A non-maximal suppression algorithm is used to filter several detection frames of each human face, and the optimal face detection frames are reserved to obtain the final face detection results.

Description

technical field [0001] The embodiments of the present application relate to the field of computer vision, and in particular to a people counting system based on video surveillance image processing. Background technique [0002] Video surveillance is the use of computer vision technology to process, analyze, and understand video signals. Without human intervention, through automatic analysis of sequence images to locate, identify, and track targets in the surveillance scene. With the development of the economy, there are more and more places for teaching, office and leisure, and the activities of the crowd are becoming more and more frequent. How to automatically detect and count the number of people in a specific area is an important issue in the field of intelligent video surveillance. and popular topics. Effectively mastering real-time population information is very important for crowd control, public space design, accident control, etc. For example, counting the number ...

AI Technical Summary

Problems solved by technology

[0003] At present, there are roughly two types of existing surveillance video people counting systems: one is to install video acquisition equipment at specific entrances and exits, to detect the targets of passing pedestrians, and to count the number of targets in continuous video over time. It is easier to achieve when there are fewer people entering and exiting, but it is powerless in an open place without a specific entrance and exit. At the same time, it is difficult to accurately count the situation with a large view and too many people.

Another type of crowd monitoring system is mainly aimed at scenes with large field of view and dense targets (hundreds or even more than a thousand people). Instead of single target detection, it does crowd density estimation based on single frame, that is, does inaccurate counting, so it cannot Target increase and decrease count in continuous video

Images