Method and apparatus for processing pictures of mobile object
View Patent ↗A feature amount of an inter-mobile unit relative movement are detected as an observation amount by an observation amount detecting section 26 , time series of the observation amounts are stored as an observation series into a storage section 27 to calculate a similarity of the observation series to a predetermined collision observation series by a classification section 28 . A determination section 29 determines to be a collision accident if, in a case where the similarity is larger than a predetermined value, a mobile unit associated with the similarity is at rest in a stoppage prohibition area set in a storage section 30 and another mobile unit is moving, and to be a mobile unit failure if collision determination conditions except for the similarity are met. By consisting of not only a first scalar obtained by quantizing a relative velocity vector between mobile units but also a second scalar obtained by quantizing a relative position vector between mobile units as the observation amount, a relative movement between mobile units is classified in more detail. A mobile unit is tracked in units of block by a mobile unit tracking section 25 to discriminate overlapped mobile units in pictures.
1. An inter-mobile unit movement classification method of processing time series pictures to classify a movement between mobile units in at least one of said pictures, comprising:
detecting, with a processor, a feature amount of a relative movement of a second mobile unit with respect to a first mobile unit as an observation amount to store a time series of said observation amounts as an observation series;
calculating, with the processor, a similarity of said observation series to a reference series, said reference series including designated previously recorded observation series of inter-mobile unit movement; and
classifying, with the processor, a movement of said second object with respect to said first mobile unit according to a value of said similarity,
wherein said observation amounts each include: a first scalar obtained by quantizing an amount associated with both a relative velocity V of said second mobile unit with respect to said first mobile unit and a distance d between said first and second mobile units; and a second scalar obtained by quantizing a relative position vector of said second mobile unit with respect to said first mobile unit; and
said first scalar is a value obtained by quantizing V·f(d), where f(d) is a monotone decreasing function of said distance d.
2. The inter-mobile unit movement classification method according to claim 1 , wherein said f(d) is expressed as k/(d+ε), where ε and k are positive constants.
3. The inter-mobile unit movement classification method according to claim 1 , wherein said similarity is a probability that said observation series arises, said probability is a value calculated on the basis of a hidden Markov model, and said reference series is a learning series for determining a model parameter.
4. An inter-mobile unit movement classification apparatus comprising:
a picture storage device storing time series pictures; and
a picture processing device processing said time series pictures to detect an anomaly of a mobile unit in at least one of said pictures,
wherein said picture processing device comprises
an observation amount detecting section detecting a feature amount of a relative movement of a second mobile unit with respect to a first mobile unit as an observation amount;
an observation series storage section storing a time series of said observation amounts as an observation series;
a similarity calculation section calculating a similarity of said observation series to a reference series, said reference series including designated previously recorded observation series of inter-mobile unit movement; and
a classification section classifying a movement of said second object with respect to said first mobile unit according to a value of said similarity;
wherein said observation amounts each include: a first scalar obtained by quantizing an amount associated with both a relative velocity V of said second mobile unit with respect to said first mobile unit and a distance d between said first and second mobile units; and a second scalar obtained by quantizing a relative position vector of said second mobile unit with respect to said first mobile unit; and
said first scalar is a value obtained by quantizing V·f(d), where f(d) is a monotone decreasing function of said distance d.
5. The inter-mobile unit movement classification apparatus according to claim 4 , wherein said f(d) is expressed as k/(d+ε), where ε and k are positive constants.
6. The inter-mobile unit movement classification apparatus according to claim 4 , wherein said similarity is a probability that said observation series arises, said probability is a value calculated on the basis of a hidden Markov model, and said reference series is a learning series for determining a model parameter.
7. A method of processing time series pictures to classify a movement between mobile units, the method comprising:
detecting, with a processor, an amount of relative movement of a second mobile unit with respect to a first mobile unit to store a time series of the detected relative movement;
calculating, with the processor, a similarity of the time series to one or more designated previously recorded reference series of inter-mobile unit movement; and
classifying, with the processor, a movement of said second object with respect to said first mobile unit according to the calculated similarity;
wherein the detected relative movement includes a scalar obtained by quantizing an amount associated with both a relative velocity V of the second mobile unit, with respect to the first mobile unit, and a distance d between the first and second mobile units, the scalar being obtained by quantizing V·f(d), where f(d) is a monotone decreasing function of said distance d.