MOBILE BODY TRACK IDENTIFICATION SYSTEM

• US 20120148102A1
• Filed: 08/13/2010
• Published: 06/14/2012
• Est. Priority Date: 08/20/2009
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Moriguchi, Yukie, Konishi, Yuusuke, Nino, Yuichi
• Granted Patent: US 8,989,438 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/30241 Trajectory
• CPC Class Codes: G06T 7/277 involving stochastic approaches, e.g. using Kalman filters

Abstract

There is provided a mobile body track identification system that determines which mobile body matches which detected track with a high precision irrespective of frequent interruption of tracks of a mobile body detected in a tracking area. Herein, hypotheses are generated by use of sets of track-coupling candidate/identification pairs, which combines track-coupling candidates, combining tracks of a mobile body detected in a predetermined time in the past, and identifications of the mobile body and which satisfies a predetermined condition. Next, identification likelihoods are calculated as likelihoods of detecting identifications in connection with tracks indicated by track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses. Identification likelihoods are integrated per each track-coupling candidate/identification pair, thus calculating an identification likelihood regarding the selected hypothesis. A most-probable hypothesis is estimated based on identification likelihoods of hypotheses.

Claims

1. A mobile body track identification system comprising:

a track-coupling candidate generation part that generates track-coupling candidates corresponding to combinations of tracks of mobile bodies detected in a predetermined time in the past;

a hypothesis generation part that generates sets of track-coupling candidate/identification pairs combining track-coupling candidates and identifications of mobile bodies detected in the predetermined time in the past, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying a predetermined condition; and

a likelihood calculation part which selects individual hypotheses so as to calculate identification likelihoods representing likelihoods of detecting identifications with respect to tracks indicated by track-coupling candidates included in each of track-coupling candidate/identification pairs ascribed to each of the selected hypotheses, and which integrates identification likelihoods per each track-coupling candidate/identification pair so as to calculate an identification likelihood per each of the selected hypotheses, thus estimating a most-probable hypothesis based on identification likelihoods of hypotheses.

2. The mobile body track identification system according to claim 1 further comprising a map storage that stores a probability map defining detection probabilities of identifications of mobile bodies in response to their positional coordinates in the tracking area in advance,

wherein the likelihood calculation part includes

an identification likelihood calculation part which specifies a position of a mobile body on tracks indicated by each track-coupling candidate at each identification detection time per each track-coupling candidate/identification pair ascribed to each of the selected hypotheses, which reads a probability value of detecting the mobile body at the position from the probability map, which calculates an identification likelihood based on the probability value, which integrates identification likelihoods per each track-coupling candidate/identification pair so as to calculate an identification likelihood regarding each of the selected hypotheses, and

a most-probable estimation part that estimates the most-probable hypothesis based on identification likelihoods of hypotheses.

3. The mobile body track identification system according to claim 2 further comprising a likelihood storage that stores track/identification correlation likelihoods representing likelihoods of correlating identifications to tracks of mobile bodies,

wherein the identification likelihood calculation part calculates a track/identification correlation likelihood, corresponding to a current combination of a track and an identification included in track-coupling candidates in track-coupling candidate/identification pairs ascribed to the most-probable hypothesis and all the hypotheses generated by the hypothesis generation part, so as to store them in the likelihood storage, and

wherein the identification likelihood calculation part calculates an identification likelihood regarding each track-coupling candidate/identification pair based on the track/identification correlation likelihood corresponding to the current combination of the track and the identification included in track-coupling candidates per each track-coupling candidate/identification pair as well as the probability value which is read from the probability map at each identification detection time.

4. The mobile body track identification system according to claim 2 further comprising a hypothesis storage that stores the most-probable hypothesis and all the hypotheses generated by the hypothesis generation part, and an information history variation determination part that determines whether or not variations occur in identifications detected in the predetermined time in the past,

wherein the track-coupling candidate generation part generates track-coupling candidates when the information history variation determination part determines that variations occur in identifications,

wherein the hypothesis generation part generates hypotheses when the information history variation determination part determines that variations occur in identifications,

wherein individual hypotheses are selected from among hypotheses generated by the hypothesis generation part when the information history variation determination part determines that variations occur in identifications, whilst individual hypotheses are selected from among hypotheses stored in the hypothesis storage when the information history variation determination part determines that no variations occur in identifications, and

wherein the most-probable hypothesis is estimated among hypotheses generated by the hypothesis generation part when the information history variation determination part determines that variations occur in identifications, so that the track-coupling candidate generation part stores the most-probable hypothesis in the hypothesis storage together with all the hypotheses generated by the hypothesis generation part.

5. The mobile body track identification system according to claim 2, wherein the hypothesis generation part correlates attributes of mobile bodies detected in the predetermined time in the past to track-coupling candidates so as to generate sets of track-coupling candidate/identification pairs, combining track-coupling candidates, correlated to attributes, and identifications of mobile bodies, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying a predetermined condition,

wherein the likelihood calculation part includes an attribute likelihood calculation part which calculates an attribute likelihood representing a likelihood of indicating the same mobile body with attributes, correlated to track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses, and which integrates attribute likelihoods per each track-coupling candidate/identification pair, thus calculating an attribute likelihood regarding each of the selected hypotheses, and

wherein the most-probable estimation part estimates the most-probable hypothesis based on identification likelihoods and attribute likelihoods of hypotheses.

6. The mobile body track identification system according to claim 2, wherein the hypothesis generation part correlates mobility values of mobile bodies detected in the predetermined time in the past to track-coupling candidates so as to generate sets of track-coupling candidate/identification pairs combining track-coupling candidates correlated to mobility values and identifications of mobile bodies, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying the predetermined condition,

wherein the likelihood calculation part includes a mobility likelihood calculation part which calculates a mobility likelihood representing a likelihood of indicating the same mobile body with mobility values correlated to track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses, and which integrates mobility likelihoods per each track-coupling candidate/identification pair, thus calculating a mobility likelihood per each of the selected hypotheses, and

wherein the most-probable estimation part estimates the most-probable hypothesis based on identification likelihoods and mobility likelihoods of hypotheses.

7. The mobile body track identification system according to claim 2, wherein the likelihood calculation part further includes a track-coupling likelihood calculation part that calculates a track-coupling likelihood representing a likelihood of connecting tracks included in each of track-coupling candidates in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses, and

wherein the most-probable estimation part estimates the most-probable hypothesis based on identification likelihoods and track-coupling likelihoods of hypotheses.

8. The mobile body track identification system according to claim 7, wherein the track-coupling likelihood calculation part calculates track-coupling likelihoods based on detection times and positional coordinates of mobile bodies defining tracks.

9. The mobile body track identification system according to claim 1, wherein the track-coupling candidate generation part calculates track-coupling likelihoods representing likelihoods of connecting tracks included in the generated track-coupling candidates so as to select track-coupling candidates whose track-coupling likelihoods are equal to or above a threshold, and

wherein the hypothesis generation part generates hypotheses by use of the selected track-coupling candidates.

10. The mobile body track identification system according to claim 1 further comprising:

an environmental information storage that stores environmental information by use of a position of obstruction which exists in a tracking area and a transient time in which a mobile body passes by the obstruction,

wherein the hypothesis generation part calculates a real transient time from a track disappearance detection time to a next track appearance detection time with respect to each track-coupling candidate, and estimates a transient time of the mobile body from a track disappearance time to a next track appearance time based on the environmental information, thus precluding track-coupling candidates in which the estimated transient time is longer than the real transient time from track-coupling candidates for use in generation of hypotheses.

11. The mobile body track identification system according to claim 1 further comprising an environmental information storage that stores a transient time of a mobile body along a passable path in a tracking area as environmental information,

wherein the hypothesis generation part calculates a real transient time from a track disappearance detection time to a next track appearance detection time with respect to each track-coupling candidate, and estimates a transient time of a mobile body from a track disappearance time to a next track appearance time based on the environmental information, thus precluding track-coupling candidates in which the estimated transient time is longer than the real transient time from track-coupling candidates for use in generation of hypotheses.

12. A mobile body track identification method comprising:

generating track-coupling candidates corresponding to combinations of tracks of mobile bodies detected in a predetermined time in the past;

generating sets of track-coupling candidate/identification pairs combining track-coupling candidates and identifications of mobile bodies detected in the predetermined time in the past, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying a predetermined condition;

selecting individual hypotheses so as to calculate identification likelihoods representing likelihoods of detecting identifications with respect to tracks indicated by track-coupling candidates included in each of track-coupling candidate/identification pairs ascribed to each of the selected hypotheses;

integrating identification likelihoods per each track-coupling candidate/identification pair so as to calculate an identification likelihood per each of the selected hypotheses; and

estimating a most-probable hypothesis based on identification likelihoods of hypotheses.

13. The mobile body track identification method according to claim 12 further comprising:

storing a probability map defining detection probabilities of identifications of mobile bodies in response to their positional coordinates in the tracking area in advance;

specifying a position of a mobile body on tracks indicated by each track-coupling candidate at each identification detection time per each track-coupling candidate/identification pair ascribed to each of the selected hypotheses;

reading a probability value of detecting the mobile body at the position from the probability map;

calculating an identification likelihood based on the probability value;

further integrating identification likelihoods per each track-coupling candidate/identification pair so as to calculate an identification likelihood regarding each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods of hypotheses.

14. The mobile body track identification method according to claim 13 further comprising:

storing track/identification correlation likelihoods representing likelihoods of correlating identifications to tracks of mobile bodies;

calculating a track/identification correlation likelihood, corresponding to a current combination of a track and an identification included in track-coupling candidates in track-coupling candidate/identification pairs ascribed to the most-probable hypothesis and all other hypotheses; and

calculating an identification likelihood regarding each track-coupling candidate/identification pair based on the track/identification correlation likelihood corresponding to the current combination of the track and the identification included in track-coupling candidates per each track-coupling candidate/identification pair as well as the probability value which is read from the probability map at each identification detection time.

15. The mobile body track identification method according to claim 13 further comprising:

storing the most-probable hypothesis and relevant hypotheses;

determining whether or not variations occur in identifications detected in the predetermined time in the past;

generating track-coupling candidates so as to generate hypotheses when it is determined that variations occur in identifications;

selecting individual hypotheses from among the generated hypotheses when it is determined that variations occur in identifications, while selecting individual hypotheses from among the stored hypotheses when it is determined that no variations occur in identifications; and

upon estimating the most-probable hypothesis among the generated hypotheses when it is determined that variations occur in identifications, storing the most-probable hypothesis together with all the other generated hypotheses.

16. The mobile body track identification method according to claim 13 further comprising:

correlating attributes of mobile bodies detected in the predetermined time in the past to track-coupling candidates so as to generate sets of track-coupling candidate/identification pairs, combining track-coupling candidates, correlated to attributes, and identifications of mobile bodies, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying a predetermined condition;

calculating an attribute likelihood representing a likelihood of indicating the same mobile body with attributes, correlated to track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses;

integrating attribute likelihoods per each track-coupling candidate/identification pair, thus calculating an attribute likelihood regarding each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods and attribute likelihoods of hypotheses.

17. The mobile body track identification method according to claim 13 further comprising:

correlating mobility values of mobile bodies detected in the predetermined time in the past to track-coupling candidates so as to generate sets of track-coupling candidate/identification pairs combining track-coupling candidates correlated to mobility values and identifications of mobile bodies, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying the predetermined condition;

calculating a mobility likelihood representing a likelihood of indicating the same mobile body with mobility values correlated to track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses;

integrating mobility likelihoods per each track-coupling candidate/identification pair, thus calculating a mobility likelihood per each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods and mobility likelihoods of hypotheses.

18. The mobile body track identification method according to claim 13 further comprising:

calculating a track-coupling likelihood representing a likelihood of connecting tracks included in each of track-coupling candidates in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods and track-coupling likelihoods of hypotheses.

19. The mobile body track identification method according to claim 18 further comprising:

calculating track-coupling likelihoods based on detection times and positional coordinates of mobile bodies defining tracks.

20. The mobile body track identification method according to claim 12 further comprising:

calculating track-coupling likelihoods representing likelihoods of connecting tracks included in track-coupling candidates so as to select track-coupling candidates whose track-coupling likelihoods are equal to or above a threshold; and

generating hypotheses by use of the selected track-coupling candidates.

21. The mobile body track identification method according to claim 12 further comprising:

storing environmental information by use of a position of obstruction which exists in a tracking area and a transient time in which a mobile body passes by the obstruction;

calculating a real transient time from a track disappearance detection time to a next track appearance detection time with respect to each track-coupling candidate;

estimating a transient time of the mobile body from a track disappearance time to a next track appearance time based on the environmental information; and

precluding track-coupling candidates in which the estimated transient time is longer than the real transient time from track-coupling candidates for use in generation of hypotheses.

22. The mobile body track identification method according to claim 12 further comprising:

storing a transient time of a mobile body along a passable path in a tracking area as environmental information;

calculating a real transient time from a track disappearance detection time to a next track appearance detection time with respect to each track-coupling candidate;

estimating a transient time of a mobile body from a track disappearance time to a next track appearance time based on the environmental information; and

precluding track-coupling candidates in which the estimated transient time is longer than the real transient time from track-coupling candidates for use in generation of hypotheses.

23. A mobile body track identification program which is loaded into and executed by a computer, said mobile body track identification program comprising:

generating track-coupling candidates corresponding to combinations of tracks of mobile bodies detected in a predetermined time in the past;

generating sets of track-coupling candidate/identification pairs combining track-coupling candidates and identifications of mobile bodies detected in the predetermined time in the past, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying a predetermined condition;

selecting individual hypotheses so as to calculate identification likelihoods representing likelihoods of detecting identifications with respect to tracks indicated by track-coupling candidates included in each of track-coupling candidate/identification pairs ascribed to each of the selected hypotheses;

integrating identification likelihoods per each track-coupling candidate/identification pair so as to calculate an identification likelihood per each of the selected hypotheses; and

estimating a most-probable hypothesis based on identification likelihoods of hypotheses.

24. The mobile body track identification program according to claim 23 further comprising:

storing a probability map defining detection probabilities of identifications of mobile bodies in response to their positional coordinates in the tracking area in advance;

specifying a position of a mobile body on tracks indicated by each track-coupling candidate at each identification detection time per each track-coupling candidate/identification pair ascribed to each of the selected hypotheses;

reading a probability value of detecting the mobile body at the position from the probability map;

calculating an identification likelihood based on the probability value;

further integrating identification likelihoods per each track-coupling candidate/identification pair so as to calculate an identification likelihood regarding each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods of hypotheses.

25. The mobile body track identification program according to claim 24 further comprising:

storing track/identification correlation likelihoods representing likelihoods of correlating identifications to tracks of mobile bodies;

calculating a track/identification correlation likelihood, corresponding to a current combination of a track and an identification included in track-coupling candidates in track-coupling candidate/identification pairs ascribed to the most-probable hypothesis and all other hypotheses; and

calculating an identification likelihood regarding each track-coupling candidate/identification pair based on the track/identification correlation likelihood corresponding to the current combination of the track and the identification included in track-coupling candidates per each track-coupling candidate/identification pair as well as the probability value which is read from the probability map at each identification detection time.

26. The mobile body track identification program according to claim 24 further comprising:

storing the most-probable hypothesis and relevant hypotheses;

determining whether or not variations occur in identifications detected in the predetermined time in the past;

generating track-coupling candidates so as to generate hypotheses when it is determined that variations occur in identifications;

selecting individual hypotheses from among the generated hypotheses when it is determined that variations occur in identifications, while selecting individual hypotheses from among the stored hypotheses when it is determined that no variations occur in identifications; and

upon estimating the most-probable hypothesis among the generated hypotheses when it is determined that variations occur in identifications, storing the most-probable hypothesis together with all the other generated hypotheses.

27. The mobile body track identification program according to claim 24 further comprising:

correlating attributes of mobile bodies detected in the predetermined time in the past to track-coupling candidates so as to generate sets of track-coupling candidate/identification pairs, combining track-coupling candidates, correlated to attributes, and identifications of mobile bodies, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying a predetermined condition;

calculating an attribute likelihood representing a likelihood of indicating the same mobile body with attributes, correlated to track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses;

integrating attribute likelihoods per each track-coupling candidate/identification pair, thus calculating an attribute likelihood regarding each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods and attribute likelihoods of hypotheses.

28. The mobile body track identification program according to claim 24 further comprising:

correlating mobility values of mobile bodies detected in the predetermined time in the past to track-coupling candidates so as to generate sets of track-coupling candidate/identification pairs combining track-coupling candidates correlated to mobility values and identifications of mobile bodies, thus generating hypotheses corresponding to sets of track-coupling candidate/identification pairs satisfying the predetermined condition;

calculating a mobility likelihood representing a likelihood of indicating the same mobile body with mobility values correlated to track-coupling candidates included in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses;

integrating mobility likelihoods per each track-coupling candidate/identification pair, thus calculating a mobility likelihood per each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods and mobility likelihoods of hypotheses.

29. The mobile body track identification program according to claim 24 further comprising:

calculating a track-coupling likelihood representing a likelihood of connecting tracks included in each of track-coupling candidates in track-coupling candidate/identification pairs ascribed to each of the selected hypotheses; and

estimating the most-probable hypothesis based on identification likelihoods and track-coupling likelihoods of hypotheses.

30. The mobile body track identification program according to claim 29 further comprising:

calculating track-coupling likelihoods based on detection times and positional coordinates of mobile bodies defining tracks.

31. The mobile body track identification program according to claim 23 further comprising:

calculating track-coupling likelihoods representing likelihoods of connecting tracks included in track-coupling candidates so as to select track-coupling candidates whose track-coupling likelihoods are equal to or above a threshold; and

generating hypotheses by use of the selected track-coupling candidates.

32. The mobile body track identification program according to claim 23 further comprising:

storing environmental information by use of a position of obstruction which exists in a tracking area and a transient time in which a mobile body passes by the obstruction;

calculating a real transient time from a track disappearance detection time to a next track appearance detection time with respect to each track-coupling candidate;

estimating a transient time of the mobile body from a track disappearance time to a next track appearance time based on the environmental information; and

precluding track-coupling candidates in which the estimated transient time is longer than the real transient time from track-coupling candidates for use in generation of hypotheses.

33. The mobile body track identification program according to claim 23 further comprising:

storing a transient time of a mobile body along a passable path in a tracking area as environmental information;

calculating a real transient time from a track disappearance detection time to a next track appearance detection time with respect to each track-coupling candidate;

estimating a transient time of a mobile body from a track disappearance time to a next track appearance time based on the environmental information; and

precluding track-coupling candidates in which the estimated transient time is longer than the real transient time from track-coupling candidates for use in generation of hypotheses.