8/12/2016 Program • 13:00h welcome and introduction (Toon Goedemé) • 13:30h Status update hardware and calibration (Bjorn Van IWT-Tetra project Tilt / Toon Goedemé) • 14:00h Status update abnormal behaviour detection (Kristof Van Engeland) • 14:45h Coffee break • 15:00h Status update person detection algorithms (Kristof Van Beeck) User group meeting • 15:45h Discussion and further planning (Joost Vennekens) December 8 th , 2016 Updated industrial users group Project abstract • Camera-based safety and security systems • Real-time reaction on incidents? Manual monitoring o Automatic processing and incident o detection • Needed components: Very reliable detection of persons 1. in camera images Reasoning system that can decide 2. if an alarm must be generated Enabling factors Project idea • State-of-the-art person detection algorithms show • Making people detection reliable, astonishing results also in difficult circumstances (fog, smoke, rain, dust, motion blur, …): o Accuracy great on standard benchmark data sets o Combine RGB and LWIR o EAVISE succeeded in running these in real-time on camera limited hardware o Adapt state-of-the-art person o Both open source and commercial-grade detection algorithms for this implementations available sensor combination • Price of LWIR-cameras descends steeply, with increasing • Use probabilistic KR for analysis resolution of situation: must an alarm be • Knowledge-representation based probabilistic reasoning generated? offers potential to analyse each situation 1
8/12/2016 People@VIPER Project goals Developing a sensor combination and software for ultra- o reliably detecting people in real-time Prof. Joost Kristof Van Dr. Floris De Dr. Kristof Prof. Toon Andy Maarten Composing a real-life reference image database for o Vennekens Engeland Smedt Van Beeck Goedemé Warrens Vandersteegen evaluating person detection techniques in difficult circumstances Studying techniques for automatic analysis of the observed o situation and classification as normal or abnormal Studying the certification procedure for camera-based o safety and security systems The demonstration and dissemination of the project o results via 5 real-life user cases Supporting industrial companies to adopt the developed o techniques in their products and services Work packages and progress Planning WP1: Hardware WP2: Person detection WP3: Alarm system 1.A Study on sensors 2.A Study on algorithms 3.A Study on AI Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 3.B Learning of ranking WP1.A: Study on sensors 1.B Hardware imple- 2.B Person detection WP1.B: Hardware realisation and calibration mentation & calibration SW implementation MP8 3.C Online learning WP1.C: Benchmark database WP2.A: Study on algoriths for person detection 1.C Benchmark 2.C Evaluation on MP6 3.D Evaluation WP2.B: Implementation algorithm person detection database Benchmark database MP7 WP2.C: Evaluation on benchmark database WP3.A: Study on learning alarm system WP3.B: Learning of ranking WP3.C: Online learning WP4: Evaluation and dissemination MP7 WP3.D: Evaluation on benchmark database MP1 MP2 MP3 MP4 MP5 WP4.A: User cases 4.B Evaluation and documentation MP7 WP4.B: Evaluation and documentation 4.A MP9 WP4.C: Certification & legal aspects User 4.C Study on certification and legal aspects WP4.D: Broad dissemination and networking MP9 Cases 4.D Broad dissimination Deliverables Master theses • Scientific publications: • Bjorn Van Tilt: “Calibratiesysteem van gecombineerde Van Beeck, Kristof, Goedemé, Toon; The Automatic Blind Spot o kleurencamera - IR sensor voor persoonsdetectie” Camera: A Vision-Based Active Alarm System; European Conference on Computer Vision Workshops; Computer Vision – • Bert Dewinter: “Ontwikkeling van een persoonsdetector op ECCV 2016 Amsterdam, The Netherlands FPGA” Van Beeck, Kristof; The automatic blind spot camera: hard real-time o detection of moving objects from a moving camera; PhD thesis KU • Nele Allaert: “Automatische detectie van patientengedrag Leuven; 2016. in hun slaapomgeving” Floris De Smedt and Toon Goedemé, How to reach top accuracy o for a visual warning system from a car, The sixth International • Mariia Zakharova: “People detection from drones with Conference on Image Processing Theory, Tools and Applications RGB+IR cameras” (IPTA'16), Oulu, Finland • Vulgarizing publications: Article in “Motion Control”: TRENDS EN ONTWIKKELINGEN BIJ o VISIESYSTEMEN; EAVISE ONDERZOEKT NIEUWE TECHNOLOGIEEN VOOR INDUSTRIELE TOEPASSINGEN 2
8/12/2016 Application cases Program • Men’s room hygienic guard (intern) • 13:00h welcome and introduction (Toon Goedemé) • Detecting people on train rails (FLIR) • 13:30h Status update hardware and calibration (Bjorn Van Tilt / Toon Goedemé) • Vehicle blind spot (GrootJebbink) • 14:00h Status update abnormal behaviour detection • Patient monitoring in bed (AlphaTronics) (Kristof Van Engeland) • Safety around agricultural vehicles (CNHi, Octinion) • 14:45h Coffee break • Safety on bridges and locks (Port Antwerp, WenZ) • 15:00h Status update person detection algorithms (Kristof • Drone surveillance (DroneMatrix) Van Beeck) • 15:45h Discussion and further planning (Joost Vennekens) Contents • Situation • Objectives Calibration of combined • Thermal camera • Calibration and mapping colour and IR-camera • Reading temperatures for pedestrian detection • Future work Bjorn Van Tilt Toon Goedemé Situation Objectives • Pedestrian detector • Using cheap thermal camera and webcam • Difficult scenario’s (night, fog, dust, …) o Seek Thermal Compact (206x156, 35°) • Adding Thermal images to detectors o Pleomax PWC-7300 Crystal Webcam (1280x1024, 35°) • Thermal camera’s getting cheaper • Reading and synchronizing cameras o But low resolutions • Calibration of both cameras o Removing lens distortion o Mapping cameras on each other o Calibration of IR to read temp • Capture different scenarios 3
8/12/2016 Seek Thermal Compact Calibration and mapping • Cheaper IR camera • Something needed to be seen in both camera’s • 206x156px • For smartphones • Board with metal circles o No Linux driver • Resistors behind every circle for heat • Using 100 frames After calibration Removed thermal Raw data Dead pixel filter images gradient B1 Finding calibration pattern Removing lens distortion Mapping both cameras Calibration and mapping From normal camera • Fit homography • Transform one world point from RGB to IR frame • Disadvantage: Uses planes => no depth To LWIR camera • Not big issue => Cameras are mounted close to each other 4
8/12/2016 Calibration and mapping Temperature readings • Never accurate. Highly depended on Average pixel error 14 material properties 12 • Hard because camera isn’t held at specific temperature 10 error [pixels] 8 6 4 2 0 1 2 3 4 5 6 7 8 Distance [m] Avg error Output Future work • Capturing different scenarios o Inside and outside o Normal daylight o Night time o During fog o During rain • Test pedestrian detection algorithm on this data Thank you for you attention! Contact: toon.goedeme@kuleuven.be Joost.vennekens@kuleuven.be 5
08/12/2016 Use Case 2: platform surveillance • motivation: people behave dangerously around railroad tracks o running around o crossing the tracks o suicide o … • but: specific impulsive behaviour is very difficult to predict! Detecting abnormal behaviour o tough to define and/or acquire “positive” training samples in public locations o movements can be very erratic o detection and tracking not always accurate enough Kristof Van Engeland – EAVISE Research scope Example video • Requirements: o distinction: “normal” � “abnormal” behaviour o (semi-)automatic operation • Questions: o how much “history” do we need to track? o desired accuracy? precision – recall? o real-time possible? Pipeline proposal Detection and Tracking • sequence (track) created for every person • person leaving and reentering screen � counts as new track • more info: see presentation by Kristof Van Beeck Preparation Classification •find people •find people •map tracks to •map tracks to •evaluate parameters •evaluate parameters regions regions •track movement •track movement •repeat training with •repeat training with •split camera view •split camera view •determine boundary •determine boundary •train model with •train model with new parameters new parameters •initialize model •initialize model between categories between categories tracks tracks •manual adjustment •manual adjustment •filter paths •filter paths Post- Detection Training processing 1
Recommend
More recommend
