Moving target identification method based on hog characteristic and system

Abstract

The invention discloses a moving target identification method based on a hog characteristic and a system. The system comprises a code book modeling target detection module for extracting a moving target region, a target association moudle confirmed by a target region as well as a moving target sorting recognition module based on the hog characteristic. The method and the system are mainly applied on an intelligent security platform. According to the invention, the moving target sorting recognition of any camera angle can be realized, and the moving target type still can be accurately recognized under the influences of different illumination conditions.

Description

technical field [0001] The invention relates to the fields of computer vision and machine learning, in particular to a method and system for recognizing moving objects based on hog features. technical background [0002] In intelligent video surveillance systems, people are most interested in monitoring moving objects in the scene. In order to provide real-time alarm and post-event target retrieval based on target type, target classification and recognition is a very important task in intelligent video surveillance systems. By identifying the category of moving targets, it is possible to more precisely define the occurrence and end of events and analyze the specific behavior of specific targets, reducing the false alarm rate of the system. [0003] However, object type recognition is a very challenging problem to meet the requirements such as the system can work in arbitrary camera angles, correctly distinguish categories under different lighting conditions and shadow effec...

AI Technical Summary

Problems solved by technology

For example, Lipton et al. used the shape characteristics of the target such as discreteness, area, and aspect ratio to divide moving targets into four categories: people, crowds, vehicles, and background interference. It is difficult to use a unified model to characterize targets with similar motion, so there are obvious deficiencies in the fine classification of targets

The recognition method based on motion characteristics is to use motion periodicity to divide moving objects into periodic moving objects and non-periodic moving objects. The effect of moving targets or non-periodic moving targets is poor, and the extraction of periodicity has certain requirements on the angle of observing the target

However, the recognition method based on the combination of two features requires high accuracy in extracting the target contour, and it will also make the overall system very complicated.

Images