Video Camera Network for Early Pest Video Camera Network for
Transcription
Video Camera Network for Early Pest Video Camera Network for
Video Camera Network for Early y Pest Detection in Greenhouses Réseau de capteurs in situ pour la détection précoce d BIOagresseurs des O g d dans lles cultures l sous S SERRE INRIA Sophia Antipolis - Méditerranée, Méditerranée PULSAR Team INRIA R Rennes - Bretagne B t Atl Atlantique, ti VISTA T Team Ch Chambre b d'A d'Agriculture i lt d des Al Alpes Maritimes, M iti CREAT R Research hC Center t INRA Avignon, UR407 Pathologie Végétale é é Objectives Research Issues A t ti real-time l ti id surveillance ill tto detect d t t and d track t k bioagressors bi g Automatic video S t early l pestt detection d t ti t allow ll i t t d pestt managementt methods th d Support to integrated and a d reduce educe pes pesticide c de use use. Develop system for in situ,, non invasive,, early p a decision support pp y y detection based on a video sensor network, network video analysis and cognitive vision. vision Motivations • In situ detection: implies introducing sensors in greenhouse. greenhouse Currently • Cognitive Cog e vision s o app approach: oac co combine e image age processing, p ocess g, machine ac e learning ea ga and d a priori knowledge, knowledge in a complete processing chain: Temperature and hygrometric conditions inside a greenhouse favor frequent and rapid attacks of bioagressor (insects, (insects acarids, acarids fungi). fungi) 1. Acquisition: Sampling objects on a large p g strategy: gy we have small,, complex p j g surface We need to define an optimal spatial sampling (INRIA & surface. CREAT) Difficult to know starting time and location of such attacks. Need to rapid N d to t identify id tif and d countt populations l ti t allow ll id decision d i i ttaking. ki Acarids b i i k traps, iin the h ffuture plant l i sensors observing sticky organs, e.g. growing stems for early detection of mature white flies. Data flow processing : intelligent acquisition process to record i l when h ti i detected. d t t d images only an iinsectt motion is Insects 2. Detection: context by , e.g. llearning 2 D t ti t t adaptation d t ti b learning l i techniques t h i i the visual appearance pp of p pests ((INRIA PULSAR). ) Spider Mite 3. Classification: adaptive p algorithms g to cope p with illumination changes g during daytime (INRIA PULSAR) + a priori knowledge of insects to detect (CREAT & INRA) INRA). Aphid Photo : Inra ((Brun)) Photo : Inra ((Brun)) 4 Tracking: dynamic environment necessitates to accommodate plant 4. movements t (INRIA VISTA) VISTA). Fungi Oïdium 5. Behavior recognition (egg laying predation): use of laying, hi h level l l scenario i a high description desc pt o language a guage (INRIA PULSAR). PULSAR) Thrips Photo : Inra ((Boissard)) Photo : Inra ((Brun)) • Botrytis Female white flyy laying its eggs Preliminary y results White fly Photo : Inra Photo : Inra ((Brun)) Prototype In Situ Acquisition Module A network video cameras ((protected t k off 5 wireless i l id t t d against g water p projection j and direct sun). ) Acquisition: sticky zoom A i iti ti k trap t In a 130 m2 greenhouse at CREAT planted with 3 i i off roses. varieties Classification: region Cl ifi ti i llabeled b l d according to insect types Expected E dR Results l A complete (steps 1. to 5.) real time video surveillance system for in situ early detection of pests in greenhouses (DIViNe). (DIViNe) Observing sticky traps continuously during d li ht daylight. High image resolution (1600x1200 pixels) at up to 10 frames per second second. Detection: region D t ti i off iinterest t t in i white A video camera on its stand observing a sticky trap Video cameras positioned uniformly in the plane horizontal plane. An extensible and reusable software architecture architecture. A cognitive vision approach (image processing and understanding, understanding machine lea ning and a priori learning, p io i knowledge) kno ledge) to p provide o ide a robust ob st and versatile e satile s system. stem In the long term: data mining for biological research to analyze bioagressor b h i behaviors. Automatic data acquisition q scheduled from distant computers Publications Towards a Video Camera Network for Early y Pest Detection in Greenhouses,, Vincent Martin Martin, Sabine Moisan (INRIA PULSAR), Bruno Paris, (INRIA-PULSAR) Paris Olivier Nicolas (CREAT) ENDURE International Conference, Conference La Grande Motte France Oct. Oct (CREAT), 2008. E l Pest P D i iin G h i and dS bi M i ICPR Early Detection Greenhouses , Vi Vincent M Martin Sabine Moisan, VAIB Workshop, Tampa USA, Dec. 2008 Funding from Provence-Alpes-Cote d’Azur d Azur Région Prototype ototype dep deployment oy e t a and d wireless e ess video deo ca camera e a network. et o J Journées é nationales ti l ARC 2008 2008, Sophia S hi Antipolis, A ti li O Octobre t b 2008