Method for realizing target identification by Faster-Vibe and classification algorithm

Abstract

The invention belongs to the field of target detection and recognition, and particularly relates to a method for achieving target recognition through Faster-Vibe and a classification algorithm. The method comprises the following steps: partitioning an input image according to pixel gray scale difference; wherein the large-gray-difference area is large in size, and the small-gray-difference area is small in size; removing areas with small gray difference; the input image may include a person, a vehicle, a small aircraft and a large aircraft; non-uniform down-sampling operation is carried out on the non-rejected area to obtain a corresponding down-sampling subarea; identifying whether the pixel of each down-sampling partition is a foreground pixel or not based on a Vibe algorithm; taking continuous foreground pixels on each down-sampling partition as a pixel set of a suspected target; processing a plurality of holes of the target according to the pixel set of the suspected target, and combining the suspected target into the same target; and inputting the pixel set of the same target into a pre-trained lightweight classification network, and determining which type of the target is a person, a vehicle, a small aircraft and a large aircraft.

Description

technical field [0001] The invention belongs to the field of target detection and recognition, in particular to a method for realizing a target recognition by Faster-Vibe and a classification algorithm. [0002] technical background [0003] In the application of airport video surveillance, real-time monitoring of specific targets in the video is required. Since the scale and angle of the monitored target will change greatly in the video, especially when multiple targets appear, relying on manual monitoring of all moving objects in the video screen will not only fail to monitor every moving object, but will also consume a lot of manpower , time and energy. [0004] Analyze the existing moving target detection algorithm: [0005] 1) The frame difference method has a large amount of calculation, and holes will appear in the image; [0006] 2) Gaussian background modeling consumes a lot of system resources. When the light changes suddenly, it is easy to cause a large area of ...

AI Technical Summary

Problems solved by technology

[0005] 1) The frame difference method has a large amount of calculation, and holes will appear in the image;

[0006] 2) Gaussian background modeling consumes a lot of system resources. When the light changes suddenly, it is easy to cause a large area of ​​false detection. When a long-term stationary object turns into motion, it is easy to produce a "ghost" phenomenon;

[0007] 3) The Vibe algorithm will have ghosting problems, shadow problems, and incomplete moving objects;

[0008] 4) The biggest defect of moving targets is that all moving objects can be detected, and the specific object attributes (people, cars, airplanes, etc.) are unknown

Images