Docstoc

Pattern Recognition and Bees

Document Sample
Pattern Recognition and Bees Powered By Docstoc
					Bio-inspired Learning for a
 Colony of Microsystems


       Rama Chellappa
           UMD
              Who am I?
• A computer vision researcher
• Interested in representation and
  recognition.
• Over the past 10 years heavy involvement
  with understanding video sequences
• A participant in the ARO MURI on Micro
  air vehicles
• Collaborations with Prof. M. Srinivasan.
      Pattern Recognition and Bees
• Behavior analysis of insects has led to advances in
  navigation, control systems etc.
• Goal: To automate the tracking and labeling of insect
  motion
   i.e., track the position and the behavior of insects.
• Joint work with Prof. M. Sreenivasan of ANU
• To appear in IEEE Trans. PAMI
                Anatomical Modeling

• All insects have similar anatomy.
• Hard Exoskeleton, soft interior.
• Three major body parts- Head, Thorax
  and abdomen.
• Each body part modeled as an ellipse.
• Anatomical modeling ensures
   – Physical limits of body parts are
       consistent.
   – Accounts for structural limitations.
   – Accounts for correlation among
       orientation of body parts
   – Insects move in the direction of
       their head.
                  Waggle Dance- 1
• Orientation of waggle axis  Direction of Food
  source.(with respect to sun).
• Intensity of waggle dance  Sweetness of food
  source.
• Frequency of waggle  Distance of food source.
• Parameters of interest in the waggle dance
   – Waggle Axis : Average orientation of Thorax during Waggle.
   – Duration of Waggle : Number of frames of waggle in each
     segment of the dance.
          Mixture Modeling for Behaviors

• Low level Motion states
   – Straight, motionless, waggle,
     turn.
• Each Behavior is a Markov
  model on such motion states.
• Switching between behaviors
  is modeled as another Markov
  Process.
• Detect frames of waggle dance
  by looking at
   – Rate of change of Abdomen
     Orientation
   – Average absolute motion of
     center of abdomen in the
     direction perpendicular to the
     axis of the bee.
       Shape, Motion and Behavior
               Encoded
              Particle Filter
• Tracking using a particle filter.
• Behavioral model in addition to motion model in
  the normal particle filter framework.
• Track both position and orientation of various body
  parts and the behavior exhibited by the bee.
• Observation model:
   – Mixture of Gaussians.
   – 5 Exemplars for the appearance of the bee.
• Maximum Likelihood estimate for both position
  and behavior.
Result
          The Grand Challenge - 1
• More and more MAVs and Microrobots will be employed for a variety
  of applications.
• Microsystems/pupil ratio will increase at an alarming rate.
• We need to figure out how a colony of such systems (less than 100)
  can organize themselves for carrying out a few well-defined tasks.
   – Landmark-based navigation
   – Terminal guidance
   – Perching
   – Surveillance
   – Looking for anomalies
• Much is known about how honeybees carry out tasks
  related to navigation, perching, hunting for honey, etc.
   – Prof. Srinivasan and several others
            The Grand Challenge – 2
• Examples of problems to be studied.
• Sensors for MAVs and Microrobots
• How to keep track of other microsystems in the colony
   – Tracking/Tagging a large number of movers in a restricted space
   – Bees are always on the move. Microsystems need to move only when
     needed.
   – Keeping an account of who went out and who came in
   – How to organize a sub-group of micro-systems for carrying out a
     specific task?
   – Activity recognition
   – Waggle dancing by MAVs!
• How to nourish microsystems?
   – Power, communication issues
   – Self-evaluation of their well being
   – Call for help.
• How will humans interact/interface with the colony?

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:5/17/2012
language:
pages:10
fanzhongqing fanzhongqing http://
About