Prev: Audio Metering implementation
Next: filter design
From: rubaina on 3 Jul 2010 14:29 Hi , I am trying to implement a particle filter for data fusion of data from 2 sensors. How will the weights be calculated for a certain state with several observations. Will each particle represent likelihoods for all sensors. Or will the particles be divided into separate divisions representing a particular sensor data, but then how will they be fused. Thanks for any insight.
From: Jerry Avins on 3 Jul 2010 15:03 On 7/3/2010 2:29 PM, rubaina wrote: > Hi , > > I am trying to implement a particle filter for data fusion of data from 2 > sensors. > How will the weights be calculated for a certain state with several > observations. > Will each particle represent likelihoods for all sensors. Or will the > particles be divided into separate divisions representing a particular > sensor data, but then how will they be fused. We may have a language problem. Is a particle a small lump of some material? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
From: Tim Wescott on 3 Jul 2010 18:32 On 07/03/2010 12:03 PM, Jerry Avins wrote: > On 7/3/2010 2:29 PM, rubaina wrote: >> Hi , >> >> I am trying to implement a particle filter for data fusion of data from 2 >> sensors. >> How will the weights be calculated for a certain state with several >> observations. >> Will each particle represent likelihoods for all sensors. Or will the >> particles be divided into separate divisions representing a particular >> sensor data, but then how will they be fused. > > We may have a language problem. Is a particle a small lump of some > material? A particle filter in this context is an extrapolation of a Kalman filter. A Kalman filter maintains a single state vector that is the most likely state or the expected value (i.e. average) of the state. A particle filter recognizes that some probability distributions are multi-modal, and maintains a number of possible state vectors, one for each anticipated hump in the probability distribution over the whole state vector's space. They're trendy, they're fun, they can absorb huge amounts of computation power... Someone will argue with me, but to some extent a Viterbi decoder is a form of particle filter, where you maintain a list of the N most likely sequences that the transmitted signal could be, and at each step of the filter you discard some and generate some new ones, always outputting from the most likely branch. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
From: Tim Wescott on 3 Jul 2010 18:35 On 07/03/2010 11:29 AM, rubaina wrote: > Hi , > > I am trying to implement a particle filter for data fusion of data from 2 > sensors. > How will the weights be calculated for a certain state with several > observations. > Will each particle represent likelihoods for all sensors. Or will the > particles be divided into separate divisions representing a particular > sensor data, but then how will they be fused. My understanding of the working of a particle filter is that each state vector candidate that you maintain is the maximum likelihood in it's own neighborhood of states. So whatever the weights are that you're talking about are, they'd be calculated from the statistics of that candidate, and from the expected statistics of the inputs. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
From: Jerry Avins on 3 Jul 2010 19:50 On 7/3/2010 6:32 PM, Tim Wescott wrote: > On 07/03/2010 12:03 PM, Jerry Avins wrote: >> On 7/3/2010 2:29 PM, rubaina wrote: >>> Hi , >>> >>> I am trying to implement a particle filter for data fusion of data >>> from 2 >>> sensors. >>> How will the weights be calculated for a certain state with several >>> observations. >>> Will each particle represent likelihoods for all sensors. Or will the >>> particles be divided into separate divisions representing a particular >>> sensor data, but then how will they be fused. >> >> We may have a language problem. Is a particle a small lump of some >> material? > > A particle filter in this context is an extrapolation of a Kalman > filter. A Kalman filter maintains a single state vector that is the most > likely state or the expected value (i.e. average) of the state. A > particle filter recognizes that some probability distributions are > multi-modal, and maintains a number of possible state vectors, one for > each anticipated hump in the probability distribution over the whole > state vector's space. > > They're trendy, they're fun, they can absorb huge amounts of computation > power... > > Someone will argue with me, but to some extent a Viterbi decoder is a > form of particle filter, where you maintain a list of the N most likely > sequences that the transmitted signal could be, and at each step of the > filter you discard some and generate some new ones, always outputting > from the most likely branch. Thank you for the education. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
|
Pages: 1 Prev: Audio Metering implementation Next: filter design |