Vision-Based Technologies for Security in Logistics Alberto Isasi aisasi@robotiker.es
INFOTECH is the Unit of ROBOTIKER-TECNALIA specialised in Research, Development and Application of Information and Communication Technologies (ICT), not only in the enterprise framework but also in the Digital Society. INFOTECH provides solutions to those Sectors in which ICT Technologies are the key for generation and/or exploitation of added value: Logistic and Industrial Sector Tourism and Entertainment. Pág. 2/67
Technological Areas Machine Vision. Semantic Technologies. Biometry. Voice Synthesis and Recognition. Trazability. Digital Ecosystems. Pág. 3/67
Application Areas Infrastructure Security Picking Access Control Industrial Control CA3 Voice based Services CA1 Image Processing Man-Machine Interfaces Audio Content Processing Visual Content Processsing CA2 Mobile Services CA4 Supply Chain Management Supply Chain Scheduling Location-based services Interoperability Unimodal & multimodal Scheduling Portable Device Services Trazability of people, machines and vehicles Pág. 4/67
Machine Vision Overview Machine Vision Overview 3D Image Processing Biometrics Pág. 5/67
Practical Machine Vision What is involved in a Machine Vision project?? Pág. 6
Practical Machine Vision So, machine vision is not just a matter of computer and image processing algorithms: Lighting and optics Acquisition system Mechanics Software Time Robustness Pág. 7
So, What is Computer Vision and Image Processing? Automatic or assisted process of information acquired by cameras Acquisition through visual resources: Hardware Tools Automatic process: Software Practical Machine Vision INFORMATION SCENE LIGHTS OPTICS HARDWARE IMAGE IN MEMORY IMAGE PROCESSING ANALYSIS RESULT Pág. 8
Practical Machine Vision Computers only see a set of numbers. It is necessary to interpret their meaning. RAW data Processing information Interpretation Pág. 9
Practical Machine Vision Colour Spaces: RGB: Colour cube. It is used by electronics devices to represent colours. HSL: Colour cone. Luminance (L) is independent from Chrominance (H+S) H Hue S Saturation L Luminance Pág. 10
Practical Machine Vision In those colour spaces (RGB, HSL ) it is no easy to measure colour similarities To create a color space that represents color information as humans see it the CIE Lab space was created 3 Parameters: Luminance, a and b Advantages: Color difference applying Euclidean Distance RGB_distance = (R1-R2)^2 + (G1-G2)^2 + (B1-B2)^2 It does not give a color similarity idea Lab_distance = (L1-L2)^2 + (a1-a2)^2 + (b1-b2)^2 It gives a color distance as humans perceive similar colors. Pág. 11
Practical Machine Vision It is possible to detect non visible characteristics for humans. Some animals as insects or deer have a ultraviolet vision that humans can t perceive Visible vs. UV Pág. 12
Near Infrared: It could be necessary an extra light source Great penetration Practical Machine Vision Mid Infrared Thermo graphic images T Beams Higher penetration Non ionizing Pág. 13
Practical Machine Vision X Beams, Gamma beams Here, it is not processed what material reflects, but what passes through the object and hits into a specific surface Due to their high energy and penetration capacity, they are ionizing Pág. 14
Spectrography: A diffraction net is used to split the light into its different wavelengths Each image pixel is defined by an N-dimension vector where each n-component represent a different wavelength. Uses: Materials, food, pathogens classification Health Aerial inspection Practical Machine Vision Spectrophotometer Hyper spectral cube One pixel spectrum Pág. 15
Image Types: Float point image Practical Machine Vision 3D Image Image sequence (video) Multidimensional image Pág. 16
Machine Vision Overview Machine Vision Overview 3D Image Processing Biometrics Pág. 17
3D Vision: Stereoscopic Stereoscopic Vision Systems: It uses 2 or more cameras. It needs that all the cameras watch the same scene Locating the same point (x, y) in each image it is possible to triangulate the 3D real Point (x, y, z) Epipolar points are calculated in calibration step Pág. 18
Stereoscopic Vision: Stereovision platforms needs to be calibrated A chessboard patter is used to calibrate the platform. It needs several images along the working area 3D Vision: Stereoscopic Stereovision platforms based in two cameras try to simulate the humans sight sense Pág. 19
Laser Beam Triangulation: 3D Vision: Laser triangulation A laser plane is projected into a region of interest The intersection between the laser and the surface of interest creates a shape This shape is shown from an external camera and is used to get the 3D information from one section of the object Pág. 20
3D Vision: Laser triangulation Laser Scan: There are differet ways to get a laser scan for surfaces: By movin the object or the set camera + laser By using a mirror to get and aditional degree of fredom moving the laser beam Pág. 21
3D Vision: Structured Light Structured Light: Projects a pattern of light on the surface and look at the deformation of the pattern on it. The pattern may be one dimensional (line) or two dimensional (grid). Needs Calibration. The pattern is projected on a calibration object. Watching how the patter is projected on the surface and comparing it with the expected deformation the system is calibrated Pág. 22
Machine Vision Overview Machine Vision Overview 3D Image Processing Biometrics Pág. 23
Biometrics: Overview What s biometrics? A technology. Can make a Relationship between a Person and a previously extracted Pattern. Allows users to Authenticate. What does this mean? You need not know anything (Passwords) You need not have to anything (Tokens) All you have to do is BE there ROBOTIKER Pág. 24
Biometrics: Overview Identification Who am I?? Peter Authentication Am I Peter? Peter DataBase Mary Peter Peter John Am I Peter? Peter Template ROBOTIKER Pág. 25
Biometrics: Definitions Several terms are used FAR (False Acceptance Rate) % of the users accepted by the system when they should not have been accepted. FRR (False Rejection Rate) % of the users who are rejected by the system when they should have been accepted. ERR (Equal Error Rate) The point in which FAR and FRR have the same value We can work at different levels of security setting a threshold to accept or to reject the doubtful cases ROBOTIKER Pág. 26
Biometrics: Fingerprint It is based on the extraction of fingerprint characteristics Minutiaes Advantage: Constant in time Good precision Disadvantages: People associate fingerprint with police It is not the most robust biometric system Pág. 27
Biometrics: Retina Feature extraction from the bottom eye s circulatory structures Eye illumination through infrared light Optic complexity User profile based on arteries and veins distribution Advantage: Excellent robustness First commercial system Disadvantages: Intrusive (too much) Information acquired from the bottom eye Pág. 28
Biometrics: Iris Feature extraction from the iris texture Eye location User profile based on texture analysis. All the algorithms for image processing are valid. Advantage: Precision Several solutions on the market Disadvantages: Intrusive Could be non collaborative There exists the Iridology: medicine based on iris analysis Pág. 29
Application to Projects Biometrics Surveillance Hyperspectral based projects Pág. 30
Biometrics: Palmprint recognition Multimodal biometric profile : Several biometric features from one measure. Easy for user: only one authentication system More robust: Different biometric measures to authenticate for each user Pág. 31
Biometrics: Palmprint recognition Palmprint Recognition prototype Computer Scanner Remote DataBase Image adquisition system TCP/IP I/O card ROBOTIKER Pág. 32
Biometrics: 2D / 3D face recognition 3D Face Model Extracted Face Points Algorithm main steps: Face orientation detection Face Points Extraction 2D Face recognition algorithm Pág. 33
Application to Projects Biometrics Surveillance Hyperspectral based projects Pág. 34
Multiview worker tracking Multicamera: One camera for each view Object detector for each camera Surveillance: worker tracking Algorithm for data coherence: fusion of the multicamera objects detection. Non collaborative RFID readers: worker identification Space coordinates transform: from camera space (3D world) to plane (2D space) System features Restricted areas intrusion detector. Abandoned object detector. Morphological detector: Humans / Trucks / Others Pág. 35
Surveillance: worker tracking Each camera has its own object detector and planespace transform algorithm With the Plant Locator Viewer it is possible to combine the different camera data with the RFID information for worker identification Pág. 36
Application to Projects Biometrics Surveillance Hyperspectral based projects Pág. 37
Hyperspectral process Use of hyperspectral images for material classification. Distinguish from same colour materials Detection of hiden objects (λ ) L r Visible Spectrum Hyperspectral pixel 250 500 750 1000 1250 Wavelength (nm) Pág. 38
Hyperspectral process Materials which seems to have the same colour can be easily distinguish analysing their hyperspectral response Pág. 39
CONCLUSIONS Pág. 40
Conclusions Image processing is a mature technology, not only a laboratory one. Biometric with image processing: It is a robust technology. It offers a high reliability. There are several different techniques. Security tasks Image processing can solve lots of surveillance issues. Image processing can be applied to a large number of applications Pág. 41
Conclusions Image Processing is more than algorithms. Optics, lights, mechanical system are at least as important as a good algorithm. Use the constrains of the application to simplify algorithms. ROBOTIKER Pág. 42
More information: aisasi@robotiker.es www.robotiker.com