Page - 21 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

SubpixelLocalisationofNanoparticles inImages ∗ ThomasHoch1,MatthiasDorfer1, andClemensHelmbrecht2 1 SCCH,Austria {thomas.hoch,matthias.dorfer}@scch.at 2 ParticleMetrixGmbH,Germany helmbrecht@particle-metrix.de Abstract Nanoparticle Tracking Analysis (NTA) is an emerging technology for the quantification of particle size, concentrationandzetapotential forparticles in thesizeregimeof10 to1000nm. The technique allowsthevisualizationof theBrownianmotionofparticles inliquidsuspensions. It is frequentlyused in commercial and academic applications for the analysis of the physical and chemical properties of dispersions suchassolubility, rheologyandreactivitywhicharestrongly influencedby thesizeof the respectiveparticles. Hence,measuring the sizeofmicro-orevennano- sizedparticles indispersions playsacentral role inchemical andbiomedical industries. With the NTA technique, particles dispersed in liquids are illuminated with an intensive light beam, e.g. from a laser. An image series of the light scattered by the particles is recorded with a sensitive digital camera with a magnification microscope attached to it. From the image series the Brownian motionof theparticles isanalyzedbyfirst localizing theparticle ineachvideoframe,secondtracking of the particles from frame to frame, and third computing the Mean Squared Displacement (MSD) along the track of each individual particle. Having the MSD one can estimate the particles diffusion coefficient and apply the Stoke-Einstein relationship to estimate the hydrodynamic size of individual particle. Current NTA systems use background segmentation method to differentiate the particles from background, mostly with fixed threshold approach. Fixed threshold works well for mono-modal dispersion since the brightness of the particles is evenly distributed. Poly-disperse particle solutions on the other hand show a high variation in the particle intensity because the reflected light intensity dependson the particle sizeand thus it isdifficult tofinda fixed threshold value. We propose a new method for NTA which utilizes a multi-scale Laplacian of a Gaussian (LoG) de- tector on top of the background-subtraction model to localize the particles. Our approach uses an optimized thresholding method for each blob individually to compute a super-resolution position es- timate. We show that our method finds more particles in the video with higher precision over the full size-range of tested solutions (20nm-500nm) in comparison to the fixed threshold approach. We furthershowthat the increasedefficiency inparticle trackingandthehigherprecision in the localiza- tionof theparticlecenter leads toparticlesizedistributions thatarenarrower(having lessvariance). Thus, our method is in particular better suited for the analysis of mixtures of poly-disperse particle solutions if the sizeof theparticles in themixture solution isnot too farapart. ∗Theresearchreported in thisarticlehasbeenpartlysupportedby theAustrianMinistry forTransport, Innovationand Technology, theFederalMinistryofScience,ResearchandEconomy, and theProvinceofUpperAustria in the frameof theCOMETcenterSCCH. 21