CN104700385A

CN104700385A - Binocular vision positioning device based on FPGA

Info

Publication number: CN104700385A
Application number: CN201310651555.8A
Authority: CN
Inventors: 胡立坤; 卢泉; 马文光; 李小为
Original assignee: Guangxi University
Current assignee: Guangxi University
Priority date: 2013-12-06
Filing date: 2013-12-06
Publication date: 2015-06-10
Anticipated expiration: 2033-12-06
Also published as: CN104700385B

Abstract

The invention discloses a binocular vision positioning device based on FPGA. The device adopts two cameras to acquire images information, adopts two SDRAM to cache the two images, and the two images are displayed in an LCD in real time. The two images are processed parallelly in the FPGA, after specific target image location information is acquired by image filtering, feature extracting and other digital image processing method, the cartesian space location information of the target object are acquired by a BP neural network calibration method. All the modules inside the FPGA are implemented through hardware programming language, the instantaneity and stability of the entire system are guaranteed, the effects caused by undetermined camera parameters and lens distortion are reduced by the BP neural network calibrating the cameras, and the positioning accuracy of the target object is improved.

Description

Based on the binocular visual positioning device that FPGA realizes

Technical field

The present invention is a kind of binocular visual positioning device realized based on FPGA, relates to machine vision technique, image processing techniques, artificial neural network technology, multiple technical field such as embedded technology.

Background technology

FPGA (Field Programmable Gate Array) is high performance programmable logic device (PLD), and can realize calculating at a high speed and parallel processing, be a kind of more excellent solution of high-performance embedded vision system.Because FPGA has the advantage of parallel processing, it exceedes traditional digital signal processor (DSP) processing speed to the computing velocity of image procossing.And based on the power consumption of FPGA embedded vision system far below the computer vision system based on universal cpu and GPU.Along with the development of technology, the integrated level of FPGA is more and more higher, and design scale is increasing, and power consumption is then more and more lower, and the embedded vision system based on FPGA will be machine vision important research direction.

In binocular visual positioning system, need the image coordinate of two video cameras to be converted to cartesian space coordinate.Traditional method needs to know the inside and outside parameter model of the video camera determined, and could obtain world coordinates position through complicated coordinate conversion.And utilize BP neural network scaling method can realize nonlinear transformation between two-dimensional signal to three-dimensional information, without the need to determining the inside and outside parameter of video camera, the impact of the various distortion of camera lens and outside noise can be reduced simultaneously, make binocular visual positioning system more accurate, stable.Vision system is mainly used in the measurement of workpiece and location in the production such as welding, carrying of industrial robot, and the target following of mobile robot, running fix etc.

Existing patent 201010185865.1 achieves real-time stereoscopic vision with FPGA, but employing is traditional Census Stereo Matching Algorithm; Patent 200810017899.2 utilizes FPGA to realize moving target and the tracking equipment of monocular.Patent 201110434839.2 realizes the sensation target self-adapting detecting controller based on FPGA, but is only limitted to detect the unrealized location of target object.

Summary of the invention

Invent a kind of binocular visual positioning device based on FPGA, whole device core is realized on FPGA by hardware program language, does not embed any soft-core processor, make system speed sooner, more stable.The core of whole device from hardware, primarily of FPGA controller [1], two CMOS camera [2], two SDRAM storage chip [3], the formation such as LCD display section [4] and RS232 communications portion [5].Whole system is controlled by FPGA [1], realize two CMOS camera [2] image acquisition and control module [6] at FPGA [1], two SDRAM storage chip [3] are carried out to the module [7] of image storage and control, image display [8], image procossing and locating module [9], BP neural network demarcating module [10] and RS232 control module [11].

During system works, the image acquisition inner by FPGA [1] and control module [6] control two CMOS camera [2], gather image information that is extraneous and target object, and image information is stored in SDRAM storage chip [3] by image storage and control module [7]; Image information carries out the rough handlings such as filtering, feature extraction, target localization by image rough handling and locating module [9] to image afterwards, obtains the image coordinate (u of target object ₁, v ₁), (u ₂, v ₂); Then two coordinate informations are input to conversion in BP neural network demarcating module [10] and obtain target object cartesian space coordinate information (x, y, z); The cartesian space coordinate information obtained outputs in other equipment such as PC, robot by RS232 control module [11] in real time; Image display [8] is controlled one piece of LCD display section [4] and shows the former digital picture that two width image acquisition and control module [6] collect or the digital picture processed by image procossing and locating module [9] simultaneously, and LCD display section [4] can show image coordinate (u in real time ₁, v ₁), (u ₂, v ₂), target object cartesian space coordinate information (x, y, z) and related status information.

Accompanying drawing explanation

Fig. 1 is the overall theory diagram based on FPGA binocular visual positioning device

In figure:

[1] fpga chip;

[2] two CMOS camera;

[3] two SDRAM chips;

[4] LCD control chip and LCD display;

[5] RS232 control chip;

[6] FPGA internal image data acquisition and controlling module;

[7] FPGA internal image stores and control module;

[8] FPGA internal image display module;

[9] process of FPGA internal image and locating module;

[10] FPGA inner BP neural network demarcating module;

[11] the inner RS232 control module of FPGA.

Embodiment

(1) utilize two CMOS camera to gather realtime graphic, utilize a LCD display to show two width realtime graphics;

(2) extract particular color target object according to Color Segmentation, calculate the image coordinate (u of target object ₁, v ₁), (u ₂, v ₂);

(3) the BP neural computing by training goes out target object cartesian space coordinate information (x, y, z);

(4) target object cartesian space coordinate information (x, y, z) is exported on other equipment by RS232 serial communication.

Claims

1. the binocular visual positioning device realized based on FPGA, this device is by FPGA controller [1], two CMOS camera [2], two SDRAM storage chip [3], the formations such as LCD display section [4] and RS232 communications portion [5], it is characterized in that using FPGA controller [1] Real-time Collection two CMOS camera [2] image informations, and realize the high-resolution image display of multi-screen with LCD display section [4]; FPGA controller [1] inside realizes image procossing and location [9], and BP neural network demarcates [10], and the final target object cartesian space positional information that obtains is transferred to external unit by R232 communications portion [5].

2. the binocular visual positioning device realized based on FPGA according to claim 1, it is characterized in that FPGA controller [1] inside realizes the parallel processing that image procossing and location [9] can realize two width images, by image median filter, Threshold segmentation, Color Segmentation, rim detection etc. method realize feature extraction and the framing of specific objective.

3. the binocular visual positioning device realized based on FPGA according to claim 1, be further characterized in that the BP neural network demarcating module [10] of FPGA controller [1] inside can not need the inside and outside parameter of video camera just can determine the nonlinear transformation relation of picture position and cartesian space position, the final target object that obtains is relative to the accurate positional information of the cartesian space of twin camera.

4. the binocular visual positioning device realized based on FPGA according to claim 1, is further characterized in that the image display [8] of FPGA controller [1] inside can realize showing the image information of two video cameras, the image location information of target object and cartesian space positional information in real time on one piece of LCD.