基于FPGA的轻量化自适应ORB算法研究与实现

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2026.01.018引用格式：，等.基于FPGA的轻量化自适应ORB算法研究与实现[J].现代电子技术，2026，49（1）：117-123.

中图分类号：TN911.73-34文献标识码：A

文章编号：1004-373X（2026）01-0117-07

Research and implementation on FPGA-based lightweight adaptive ORBalgorithm

WANGDingxuan，YAORongbin，ZHAO Zhonghua，LI Xiaohuan （Schoolof Informationand Communication，Guilin Universityof Electronic Technology，Guilin541oo4，China）

Abstract：Inorder toreducecomputationcomplexityandimprovethereal-timeperformanceofORBalgorithm，andgetrid ofthefacthatthealgorithmfilstodetectthefeaturesinthelightchangeandlowtexturescensduetoitsfixedthresholda lightweightadaptiveORBalgorithmaccelerationarchitecturebasedonFPGAisproposed.Firstly，thefeaturedirection calculationoftheORBalgorithmisimprovedbyadoptingaregion-basedfeaturedirectionangleanddescriptorcalculation method，whichreduces teconsumptionofcomputingresources.Incombinationwiththeparalelizationandpipelinecomputing advantagesofFPGA，alightweightORBaccelerationarchitectureisdesigned.Secondly，ahistogramequalizationalgorithmis addedonthebasisof theoriginalalgorithmtoadjustthebrightnessoftheimage，improvethecontrastoftheimage，andmake thefeaturedetailsoftheimagemoreprominent.Anadaptivethresholdcalculationmethodisdesignedforthefixedthresholdof theORBalgorithm，which increasesthenumberoffeature pointsextractedbythealgorithm inweak lightingandlowtexture scenes.Teexperimentalresultsshowthatincomparisonwiththesoftwarealgorithmimplementation，thehardwareacceleration architecturebasedonFPGAcanachievea16.1timesaccelerationefect.Thenumberoffeature pointsextractedunderweak lightingandlowtextureconditionsis6.67timesand2.56timesthatofORBalgorithm，respectively，andthenumberoffeature matchingpointpairsis5.62timesand1.5timesthatofORBalgorithm，respectively.Itcanbeseenthatthealgorithm acceleration isachievedandresourceconsumptionisreduced，andtheadaptabilityandtherobustnessofthealgorithmin different scenarios are enhanced.

Keywords：ORB;featuredetection;FPGA;lightweight;histogram equalization;adaptive threshold; weak lighting;lowtexture

0 引言

ORB（OrientatedFASTandRotatedBRIEF）算法是一种具有代表性的特征点提取算法，具有良好的旋转和尺度不变性，被广泛应用在VSLAM、三维重建、图像拼接等场景，但这些应用大多是无人机、机器人、自动驾驶等平台，而ORB算法的计算量对于这些应用仍然很大，无法满足实时性和低功耗的要求；与此同时，传统的ORB算法固定阈值，在光照充足和纹理特征强的场景适用，而在弱光照或低纹理特征的场景下，无法保证提取到足够多的特征点，难以真实反映图像的特征信息，导致后续的匹配结果不理想。（剩余8016字）