Image processing method of Gaussian mixture model based on FPGA

Abstract

The invention relates to an image processing method of a Gaussian mixture model based on an FPGA (Field Programmable Gate Array). The method comprises the following steps: S1, constructing FPGA module design and initial configuration of an ARM (Advanced RISC Machines) system; s2, converting pixels on the DDR and background model parameters into data streams by a group of DMA controllers, and transmitting the data streams into a core calculation module of a Gaussian mixture model; s3, a core calculation module of the Gaussian mixture model calculates that each pixel belongs to a background or a foreground, and then the parameters of the background model are updated; and S4, the other group of DMA controllers store the calculation result of the core calculation module and the updated background model parameters on the DDR. The method is used for carrying out a moving target detection task in an edge calculation scene, the hardware characteristics of the FPGA and the calculation mode of the Gaussian mixture model are combined, the performance of the Gaussian mixture model on FPGA hardware is improved through a software and hardware collaborative optimization method, and a higher processing speed is obtained under the condition that fewer FPGA on-chip resources are occupied.

Description

technical field [0001] The invention belongs to the technical field of real-time video analysis, and in particular relates to an image processing method based on an FPGA-based mixed Gaussian model. Background technique [0002] With the development of information technology, massive amounts of new data are being generated all the time in the world, and many applications of intelligent technologies rely on the calculation and processing of these data. The traditional method adopts a centralized computing mode, and the data generated at the edge location is transmitted to the central server for calculation through the network, and then the calculation result is sent back to the edge end waiting for the result through the network. However, with the continuous growth of the amount of data, the network is gradually unable to undertake such a huge data transmission task. In scenarios with high real-time requirements such as traffic detection and security monitoring, the network tr...

AI Technical Summary

Problems solved by technology

[0005] However, when the existing mixed Gaussian model is implemented on FPGA, there are the following problems in terms of performance and resource occupation: because the mixed Gaussian model needs to establish and maintain a corresponding background model for each pixel in the image, resulting in a large amount of calculation , and the on-chip memory of the FPGA is not enough to store the background model parameters at high resolution, so the background extraction method of the mixed Gaussian model on the FPGA has bottlenecks in computing performance and memory bandwidth, and it does not perform well in terms of processing speed and resource occupation. Good, but cannot meet the growing demand for real-time intelligent analysis of high-definition video

Images