Multi-camera non-overlapping view field-based pedestrian matching method

Abstract

The invention discloses a multi-camera non-overlapping view field-based pedestrian matching method, which comprises the following steps of: establishing a hybrid Gaussian background model; obtaining a pedestrian object by using a background subtraction method; extracting the main color characteristic of the pedestrian object; performing main color judgment analysis; extracting the spatial texture characteristic of a suspected matched object; and performing further judgment by using the spatial texture characteristic. The limitation of the conventional single-camera view field is broken, so that a monitoring range is effectively widened, a camera dead zone of large-range monitoring is eliminated, and an object which is separated in time and space is matched in different cameras. The method is high in robustness for the posture change of the pedestrian object and the illumination change of an environment.

Description

technical field [0001] The invention belongs to the field of computer vision, specifically relates to the field of intelligent monitoring, in particular to a pedestrian matching method based on multi-camera non-overlapping view fields. Background technique [0002] With the development of computer technology and the continuous improvement of people's safety awareness, people pay more and more attention to video surveillance technology. Due to the limited field of view of the traditional single camera and the small monitoring range, multiple cameras are often used for large-scale monitoring in most occasions, such as security monitoring inside communities and important buildings, monitoring of urban road network traffic conditions, subway stations and airports monitoring, as well as the monitoring of parking lots and supermarkets, etc. However, there must be camera blind spots in large-scale monitoring, and the targets are separated in time and space. How to match the same t...

AI Technical Summary

Problems solved by technology

However, due to the variety of targets that may appear in the monitoring scene, the posture of the target will also change during the movement process, so that the target shows various shapes, and it is difficult to establish a unified model to describe it.

At the same time, the model-based matching method relies too much on the target model. Once the model matching fails, the moving target cannot be correctly identified. Although the establishment of a fine target model can improve the matching accuracy, it will inevitably lead to an increase in the amount of calculations in the matching process. Multiplied, difficult to meet real-time requirements

Images