# US 2019/0318273 A1 · System and Method For Single-Loop Vehicle Classification Using Inductive Loop Vehicle Signatures

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0318273 A1
- Document ID: US20190318273A1
- Status: Published application
- Kind: A1
- Application number: 15/955,647
- Filed: 2018-04-17
- Published: 2019-10-17
- Earliest claimed: 2018-04-17
- Assignees: CLR Analytics Inc.
- Applicants: CLR Analytics Inc.
- Inventors: Shin-Ting Jeng; Qi Lin; Huabing Wang; Lianyu Chu
- Web page: https://patentgrep.com/patent/US20190318273A1

## CPC

- G08G 1/015: with provision for distinguishing between two or more types of vehicles, e.g. between motor-cars and cycles
- G06F 18/2413: based on distances to training or reference patterns
- G06N 5/047: Pattern matching networks; Rete networks
- G06N 20/00: Machine learning
- G08G 1/042: using inductive or magnetic detectors
- G08G: TRAFFIC CONTROL SYSTEMS (guiding railway traffic, ensuring the safety of railway traffic B61L; arrangement of road signs or traffic signals E01F9/00; radar or analogous systems, sonar systems, lidar systems specially adapted for traffic control G01S13/91, G01S15/88, G01S17/88; {radar or analogous systems, sonar systems, lidar systems specially adapted for anti-collision purposes G01S13/93, G01S15/93, G01S17/93})
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Single loop inductive sensors are widely deployed in infrastructure for traffic data collection, however, these loops currently provide little more than vehicle detection. A system and method are provided that enable single loop inductive sensors to be used for vehicle classification (e.g., identification as motorcycle, passenger car, bus, etc.). Classification may be done using the Federal Highway Administration's 13 class system. Initially a signature library is built from vehicle signatures with known classifications. Vehicle signature waveforms of unknown classification obtained from inductive loop sensors are analyzed to identify specific features in the waveform including the number of “peaks”, the first peak location and its magnitude. A classifier (e.g., K-nearest neighbor) uses a representation of the vehicle signature and the features to determine from the signature library the classification of the vehicle.

## Claims

### Claim 1

A system for classifying vehicles, the system comprising:
a non-transitory computer readable memory configured to store a classification library having a plurality of entries, each entry in the library having a vehicle classification;
a sequence generating module operably connected to an inductive loop sensor for receiving a vehicle signature therefrom, and configured to convert the vehicle signature into a sequence of numbers;
an input set generating module operably connected to the sequence generating module for receiving the sequence, and configured to identify peaks in the sequence and to generate an input set, the input set comprising the sequence, a quantity of peaks in the sequence, a location of a first peak in the sequence, and a value of the first peak;
a classification module operably connected to the memory for accessing the classification library, operably connected to the input set generating module for receiving the input set, and configured to perform K-nearest neighbor (KNN) classification with the classification library on the input set thereby determining an estimated vehicle classification for the input set; and
an output module configured to output the estimated vehicle classification.

### Claim 2 (depends on 1)

The system of claim 1, wherein the output module records the estimated vehicle classification to the memory along with the vehicle signature and station data.

### Claim 3 (depends on 1)

The system of claim 1, wherein the vehicle classification of each entry in the vehicle signature library is one of 13 Federal Highway Administration classifications.

### Claim 4 (depends on 1)

The system of claim 1, wherein the sequence generating module calculates wavelet coefficients for the vehicle signature and forms the sequence by regenerating the vehicle signature using a subset of the wavelet coefficients.

### Claim 5 (depends on 4)

The system of claim 4, wherein the sequence generating module uses the largest N % of the wavelet coefficients to regenerate the vehicle signature, where N is in the interval [50, 100).

### Claim 6 (depends on 5)

The system of claim 5, wherein the wavelet coefficients calculated by the sequence generating module are Haar wavelet coefficients, and the sequence has 32 numbers.

### Claim 7 (depends on 1)

The system of claim 1, wherein the input set generating module identifies as a peak each number in the sequence where
an immediately preceding number is less than said number and an immediately following number is less than said number (+,−), or
the immediately preceding number is less than said number and the immediately following number is equal to said number (+,0), or
the immediately preceding number is greater than said number and the immediately following number is greater than said number (−,+), or
the immediately preceding number is equal to said number and the immediately following number is greater than said number (0,+), or
the immediately preceding number is greater than said number and the immediately following number is equal to said number (−,0), or
the immediately preceding number is equal to said number and the immediately following number is less than said number (0,−).

### Claim 8 (depends on 7)

The system of claim 7, wherein the input set generating module identifies the location in the sequence of the first peak as an index of a first number in the sequence identified as a peak, and identifies the value of the first peak as said first number.

### Claim 9 (depends on 1)

The system of claim 1, wherein the input set generating module normalizes each number in the input set.

### Claim 10 (depends on 9)

The system of claim 9, wherein the classification module calculates a Euclidian distance between the input set and each of the plurality of entries in the classification library, identifies K (K a positive integer) smallest Euclidian distance entries, and determines the estimated vehicle classification based on said K entries.

### Claim 11

A system for generating a vehicle classification library, the system comprising:
a non-transitory computer readable memory storing a plurality of training examples, each training example having a vehicle signature from an inductive loop sensor and a vehicle classification;
a grouping module configured to group training examples having a same vehicle classification;
a generating module to generate for each group a set of inputs by, for each training example in the group, (i) converting the vehicle signature into a sequence, (ii) identifying peaks in the sequence, and (iii) forming an input containing (A) a quantity of peaks in the sequence, (B) a location of a first peak in the sequence, (C) a value of the first peak, and (D) values of said sequence at select peak locations;
a clustering module configured to receive the set of inputs for a group from the generating module and use K-means clustering to organize the inputs of said set into K clusters;
a library module configured to create a library entry for each cluster by (i) averaging the value of the location of the first peak among the inputs in said cluster, (ii) averaging the value of the first peak among the inputs in said cluster, (iii) averaging the value of the quantity of peaks among the inputs in said cluster, and (iv) averaging the sequences corresponding to the inputs in said cluster at each position in said sequences; and
a recording module to add each library entry created by the library module to the vehicle classification library.

### Claim 12 (depends on 11)

The system of claim 11, wherein the generating module identifies as a peak each number in the sequence where
an immediately preceding number is less than said number and an immediately following number is less than said number (+,−), or
the immediately preceding number is less than said number and the immediately following number is equal to said number (+,0), or
the immediately preceding number is greater than said number and the immediately following number is greater than said number (−,+), or
the immediately preceding number is equal to said number and the immediately following number is greater than said number (0,+), or
the immediately preceding number is greater than said number and the immediately following number is equal to said number (−,0), or
the immediately preceding number is equal to said number and the immediately following number is less than said number (0,−).

### Claim 13 (depends on 11)

The system of claim 11 wherein, to convert the vehicle signature into the sequence, the generating module is configured to calculate wavelet coefficients for the vehicle signature and regenerate the vehicle signature using a subset of the wavelet coefficients.

### Claim 14 (depends on 13)

The system of claim 13, wherein the subset of the wavelet coefficients used are the largest N % of the wavelet coefficients, where N is in the interval [50, 100).

### Claim 15 (depends on 14)

The system of claim 14, wherein the wavelet coefficients are Haar wavelet coefficients, and the sequence has 32 numbers.

### Claim 16 (depends on 11)

The system of claim 11, wherein the select peak locations are the 4 or more most frequent peak locations for vehicle signatures within the group.

### Claim 17 (depends on 11)

The system of claim 11 wherein
each group is for one Federal Highway Administration vehicle classification, and
the select peak locations are the 4 or more most frequent peak locations for vehicle signatures having a same CLRVC.

### Claim 18 (depends on 11)

The system of claim 11, wherein the recording module is further configured to store the vehicle classification library in the memory, and the system further comprises:
a sequence generating module operably connected to an inductive loop sensor for receiving an unclassified vehicle signature therefrom, and configured to convert the unclassified vehicle signature into an unclassified sequence;
an unclassified input set generating module operably connected to the sequence generating module for receiving the unclassified sequence, and configured to identify peaks in the unclassified sequence and to generate an unclassified input set, the unclassified input set comprising the unclassified sequence, a quantity of peaks in the unclassified sequence, a location of a first peak in the unclassified sequence, and a value of the first peak in the unclassified sequence;
a classification module operably connected to the memory for accessing the vehicle classification library, operably connected to the unclassified input set generating module for receiving the unclassified input set, and configured to perform K-nearest neighbor (KNN) classification with the vehicle classification library on the unclassified input set thereby determining an estimated vehicle classification for the unclassified input set; and
an output module configured to output the estimated vehicle classification.

### Claim 19

A method for generating a vehicle classification library, the method comprising:
receiving a plurality of training examples, each training example having a vehicle signature from an inductive loop sensor and a vehicle classification;
grouping the training examples having a same vehicle classification; and
for each group
generating a set of library inputs by, for each of the training examples in the group,
converting the vehicle signature into a sequence;
identifying peaks in the sequence;
calculating a library input comprising a quantity of peaks in the sequence, a location of a first peak in the sequence, a value of the first peak, and values of said sequence at select peak locations;
clustering library inputs in the set of library inputs into K clusters using a K-means clustering;
creating a library entry for each cluster by
averaging the value of the location of the first peak among the library inputs in said cluster;
averaging the value of the first peak among the library inputs in said cluster;
averaging the value of the number of peaks among the library inputs in said cluster; and
averaging the sequences corresponding to the library inputs in said cluster at each position in the sequences; and
adding the library entry to a vehicle classification library.

### Claim 20 (depends on 19)

The method of claim 19 further comprising:
receiving an unclassified vehicle signature from an inductive loop sensor;
representing the unclassified signature as an unclassified sequence;
identifying peaks in the unclassified sequence;
performing a K-nearest neighbor (KNN) classification with the vehicle classification library on an input set to determine an estimated vehicle classification for the unclassified vehicle signature, the input set comprising the unclassified sequence, a quantity of peaks in the unclassified sequence, a location of a first peak in the unclassified sequence, and a value of the first peak; and
storing the estimated vehicle classification with the unclassified vehicle signature and station data in a non-transient, computer-readable memory device.