Vehicle classification from laser scanners using fisher and profile signatures

• US 9,683,836 B2
• Filed: 08/09/2013
• Issued: 06/20/2017
• Est. Priority Date: 08/09/2013
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Inventors: Csurka, Gabriela, Poirier, Herve, Sandhawalia, Harsimrat Singh, Rodriguez Serrano, Jose Antonio
• Primary Examiner(s): Teixeira Moffat, Jonathan C
• Assistant Examiner(s): Dinh, Lynda
• Application Number:
  US13/963,472
• Publication Number:
  US 20150046119A1
• Time in Patent Office: 1,411 Days
• Field of Search: 702167
• CPC Class Codes: G01B 11/24 for measuring contours or curvatures
• CPC Class Codes: G01B 11/2518 Projection by scanning of the object
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: G06V 10/464 using a plurality of salient features, e.g. bag-of-words [BoW] representations
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
• CPC Class Codes: G06V 20/653 by matching three-dimensional models, e.g. conformal mapping of Riemann surfaces
• CPC Class Codes: G06V 20/695 Preprocessing, e.g. image segmentation
• CPC Class Codes: G06V 20/698 Matching; Classification
• CPC Class Codes: G06V 40/382 Preprocessing; Feature extraction

Abstract

Methods, systems and processor-readable media for vehicle classification. In general, one or more vehicles can be scanned utilizing a laser scanner to compile data indicative of an optical profile of the vehicle(s). The optical profile associated with the vehicle(s) is then pre-processed. Particular features are extracted from the optical profile following pre-processing of the optical profile. The vehicle(s) can be then classified based on the particular features extracted from the optical feature. A segmented laser profile is treated as an image and profile features that integrate the signal in one of the two directions of the image and Fisher vectors which aggregate statistics of local “patches” of the image are computed and utilized as part of the extraction and classification process.

Claims

1. A method for vehicle classification utilizing a laser scanner, said method comprising:

scanning at least one vehicle with a laser scanner comprising a time-of-flight measuring device that emits short pulses of light toward said at least one vehicle to generate scanned data, said scanning is repeated in a varying set of angles and each scan covers a linear row up to 100 degrees, wherein a number of rows depending on the length and/or speed of the vehicle;

compiling with a processor data indicative of an optical profile of said at least one vehicle from said scanned data generated from said laser scanner;

signal pre-processing with said processor said optical profile associated with said at least one vehicle, wherein said optical profile comprises a laser profile that is pre-processed with said processor wherein an input for all feature descriptors comprises a 2D cropped laser signal corresponding to a largest component in an original laser profile with respect to said laser profile;

extracting with said processor particular features from said optical profile after preprocessing said optical profile, said particular features including at least two attribute features extracted by segmenting said laser profile into units with threshold heuristics; and

classifying with at least one trained classifier said at least one vehicle based on said particular features extracted from said optical profile, said at least one trained classifier comprising a one-versus all linear classifier.

2. The method of claim 1 wherein said signal pre-processing with said processor said optical profile associated with said at least one vehicle, further comprises:

isolating at least one individual profile from said data indicative of said optical profile;

correcting scan reading errors associated with scanning said at least one vehicle utilizing said laser scanner; and

segmenting data indicative of a vehicle body with respect to said at least one vehicle from said data indicative of said optical profile.

3. The method of claim 1 wherein extracting with said processor particular features from said optical profile after said signal pre-processing said optical profile, further comprises:

extracting said particular features including Fisher vectors as a part of a Fisher framework to build Fisher laser signatures, said particular features including a set of local patches comprising sub-regions of at least one Fisher laser signature among said Fisher laser signatures, wherein said local patches are extracted from a regular grid of coordinates in different scales.

4. The method of claim 3 wherein extracting with said processor said particular features from said optical profile after signal pre-processing said optical profile, further comprises:

extracting said particular features including profile features by computing an integral in a horizontal direction and in a vertical direction with respect to said optical profile.

5. The method of claim 4 wherein classifying with said at least one trained linear classifier said at least one vehicle based on said particular features extracted from said optical feature, further comprises:

training said at least one linear classifier utilizing said particular features extracted from said optical feature to produce said at least one trained linear classifier; and

classifying said at least one vehicle based on data output from said at least one trained linear classifier.

6. The method of claim 5 further comprising configuring said at least one trained linear classifier to comprise an SVM (Support Vector Machine) trained via SGD (Stochastic Gradient Descent) and wherein said at least two attribute features are selected from among a plurality of attribute features extracted from said laser profile, said plurality of attribute features including a width of said laser profile, a length of said laser profile, a maximum height of said laser profile, a number of units, and a maximum height of said first unit, and wherein said laser profile comprises a 2D laser profile.

7. The method of claim 5 wherein said particular features extracted from said optical feature includes human-used attributes and said at least one trained classifier comprises a classifier trained by machine learning.

8. The method of claim 5 wherein:

said particular features extracted from said optical feature include a feature set extracted from profile images that include;

raw profile features;

Fisher image signatures; and

Fisher laser signatures; and

said laser scanner is located on a gantry in a road having a plurality of lanes, said laser scanner having a laser curtain facing downwards and perpendicular to a direction of traffic so that said laser scanner measures a height of said at least one vehicle in at least on transversal slice.

9. A system for vehicle classification utilizing a laser scanner, said system comprising:

at least one laser scanner comprising a time-of-flight measuring device that emits short pulses of light toward said at least one vehicle to generate scanned data, said scanning is repeated in a varying set of angles and each scan covers a linear row up to 100 degrees, wherein a number of rows depending on the length and/or speed of the vehicle;

at least one processor that communicates with said at least one laser scanner; and

a non-transitory computer-usable medium embodying computer program code, said non-transitory computer-usable medium communicating with said at least one processor, said computer program code comprising instructions executable by said at least one processor and configured for;

scanning at least one vehicle with said at least one laser scanner to compile data indicative of an optical profile of said at least one vehicle;

signal pre-processing said optical profile associated with said at least one vehicle, wherein said optical profile comprises a laser profile that is pre-processed with said processor wherein an input for all feature descriptors comprises a 2D cropped laser signal corresponding to a largest component in an original laser profile with respect to said laser profile;

extracting particular features from said optical profile after preprocessing said optical profile, said particular features including at least two attribute features extracted by segmenting said laser profile into units with threshold heuristics; and

classifying with at least one trained classifier said at least one vehicle based on said particular features extracted from said optical profile, said at least one trained classifier comprising a one-versus all linear classifier.

10. The system of claim 9 wherein said instructions for signal pre-processing with said at least one processor said optical profile associated with said at least one vehicle, further comprise instructions configured for:

isolating at least one individual profile from said data indicative of said optical profile;

correcting scan reading errors associated with scanning said at least one vehicle utilizing said laser scanner; and

segmenting data indicative of a vehicle body of said at least one vehicle from said data indicative of said optical profile.

11. The system of claim 9 wherein said instructions for extracting particular features from said optical profile after said signal pre-processing d opt profile, further comprise instructions configured for:

extracting said particular features including Fisher vectors as a part of a Fisher framework to build Fisher laser signatures that are square-rooted and L2-normalized.

12. The system of claim 9 wherein said instructions for extracting said particular features from said optical profile after said signal pre-processing said optical profile, further comprise instructions configured for:

extracting said particular features including profile features by computing an integral in a horizontal direction and in a vertical direction with respect to said optical profile, said at least one laser scanner located on a gantry in a road having a plurality of lanes, said at least one laser scanner having a laser curtain facing downwards and perpendicular to a direction of traffic so that said at least one laser scanner measures a height of said at least one vehicle in at least on transversal slice.

13. The system of claim 12 wherein said instructions for classifying said at least one vehicle based on said particular features extracted from said optical feature, further comprise instructions configured for:

training at least one linear classifier utilizing said particular features extracted from said optical feature, said at least one trained classifier comprising said at least one linear classifier; and

classifying said at least one vehicle based on data output from said at least one linear classifier.

14. The system of claim 13 wherein said instructions are further modified for configuring said linear classifier to comprise an SVM (Support Vector Machine) trained via SGD (Stochastic Gradient Descent).

15. The system of claim 13 wherein said particular features extracted from said optical feature includes human-used attributes.

16. The system of claim 13 wherein said particular features extracted from said optical feature include a feature set extracted from profile images that include:

raw profile features;

Fisher image signatures; and

Fisher laser signatures.

17. A non-transitory processor-readable medium storing code representing instructions to cause a process for vehicle classification utilizing a laser scanner, said code comprising code to:

scan at least one vehicle utilizing a laser scanner to compile data indicative of an optical profile of said at least one vehicle, said scanning is repeated in a varying set of angles and each scan covers a linear row up to 100 degrees, wherein a number of rows depending on the length and/or speed of the vehicle;

pre-process said optical profile associated with said at least one vehicle, wherein said optical profile comprises a laser profile that is preprocessed with said processor wherein an input for all feature descriptors comprises a 2D cropped laser signal corresponding to a largest component in an original laser profile with respect to said laser profile;

extract particular features from said optical profile after preprocessing said optical profile, said particular features including at least two attribute features extracted by segmenting said laser profile into units with threshold heuristics; and

classify said at least one vehicle with at least one trained classifier based on said particular features extracted from said optical feature, said at least one trained classifier comprising a one-versus all linear classifier.

18. The non-transitory processor-readable medium of claim 17 wherein said code to pre-process said optical profile associated with said at least one vehicle, further comprises code to:

isolate at least one individual profile from said data indicative of said optical profile;

correct scan reading errors associated scanning said at least one vehicle utilizing said laser scanner; and

segment data indicative of a vehicle body of said at least one vehicle from said data indicative of said optical profile.

19. The non-transitory processor-readable medium of claim 17 wherein said code to extract particular features from said optical profile after preprocessing said optical profile, further comprises code to:

extract said particular features including Fisher vectors, said particular features including a set of local patches comprising sub-regions of at least one Fisher laser signature, wherein said local patches are extracted from a regular grid of coordinates in different scales.

20. The non-transitory processor-readable medium of claim 17 wherein said code to extract said particular features from said optical profile after preprocessing said optical profile, further comprises code to:

extract said particular features including profile features by computing an integral in a horizontal direction and in a vertical direction with respect to said optical profile, wherein said particular features include at least two attribute features selected from among a plurality of attribute features extracted from said laser profile, said plurality of attribute features including a width of said laser profile, a length of said laser profile, a maximum height of said laser profile, a number of units, and a maximum height of said first unit, and wherein said laser profile comprises a 2D laser profile.