# US 2026/0097787 A1 · Unified matching method and unified matching system for multi-pedestrian tracking

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

- Number: US 2026/0097787 A1
- Document ID: US20260097787A1
- Status: Published application
- Kind: A1
- Application number: 19/347,282
- Filed: 2025-10-01
- Published: 2026-04-09
- Earliest claimed: 2024-10-07
- Assignees: Hl Klemove Corp.
- Applicants: HL Klemove Corp.
- Inventors: Hyeonchul JUNG; Seokjun KANG; Takgen KIM; Hyeongki KIM
- Web page: https://patentgrep.com/patent/US20260097787A1

## CPC

- B60W 60/001: {Planning or execution of driving tasks}
- G06T 7/277: involving stochastic approaches, e.g. using Kalman filters
- G06V 10/74: Image or video pattern matching; Proximity measures in feature spaces
- G06V 20/58: Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V 40/103: {Static body considered as a whole, e.g. static pedestrian or occupant recognition}
- B60W 2554/40: Dynamic objects, e.g. animals, windblown objects
- B60W: CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A unified matching method for multi-pedestrian tracking according to an embodiment of the present disclosure comprises: dividing road environment images collected from at least one sensor into frames and detecting an object in an N-th frame; calculating a matching cost matrix between the detected object and tracked objects detected in a previous frame of the road environment images; matching the detected object with one of the tracked objects using the matching cost matrix; generating a matching threshold value based on a confidence score of the detected object; updating the matched tracked object using the detected object when a matching cost between the matched tracked object and the detected object is smaller than the matching threshold value; and controlling a vehicle based on the updated tracked object.

## Claims

### Claim 1

A unified matching method for multi-pedestrian tracking, the method comprising:
dividing road environment images collected from at least one sensor into frames and detecting an object in an N-th frame;
calculating a matching cost matrix between the detected object and tracked objects detected in a previous frame of the road environment images;
matching the detected object with one of the tracked objects using the matching cost matrix;
generating a matching threshold value based on a confidence score of the detected object;
updating the matched tracked object using the detected object when a matching cost between the matched tracked object and the detected object is smaller than the matching threshold value; and
controlling a vehicle based on the updated tracked object.

### Claim 2 (depends on 1)

The unified matching method of claim 1, wherein the generating of the matching threshold value comprises:
determining the confidence score of the detected object; and
generating the matching threshold value by multiplying the confidence score with a predetermined threshold constant value.

### Claim 3 (depends on 2)

The unified matching method of claim 2, wherein the generating of the matching threshold value further comprises generating a matching threshold matrix including matching threshold values, and
wherein if the number of the tracked objects is n and the number of detected objects is m, a dimension of the matching threshold matrix is (n, m).

### Claim 4 (depends on 1)

The unified matching method of claim 1, wherein the calculating of the matching cost matrix comprises:
predicting expected positions of the tracked objects in the N-th frame by using a Kalman filter; and
calculating the matching cost matrix based on Intersection over Union (IoU) values between bounding boxes of the tracked objects at the expected positions and a bounding box of the detected object.

### Claim 5 (depends on 4)

The unified matching method of claim 4, wherein the calculating of the matching cost matrix further comprises:
calculating embedding vectors of the tracked objects by inputting regions of the bounding boxes of the tracked objects into a feature extractor;
calculating an embedding vector of the detected object by inputting a region of the bounding box of the detected object into the feature extractor; and
calculating the matching cost matrix based on cosine distances between the embedding vectors of the tracked objects and the embedding vector of the detected object.

### Claim 6 (depends on 1)

The unified matching method of claim 1, wherein the matching of the detected object with one of the tracked objects comprises:
performing one-to-one matching between the detected object and the tracked objects by using a Hungarian algorithm.

### Claim 7 (depends on 1)

The unified matching method of claim 1, wherein if the confidence score of the detected object is greater than or equal to a predetermined threshold value and the detected object is not matched with any one of the tracked objects, a new tracked object is generated by using the detected object.

### Claim 8 (depends on 7)

The unified matching method of claim 7, wherein the new tracked object is deleted when the new tracked object is not matched with an object detected in a subsequent frame.

### Claim 9 (depends on 7)

The unified matching method of claim 7, wherein the new tracked object becomes a confirmed tracked object when the new tracked object is subsequently matched for at least a predetermined number of consecutive frames, and the confirmed tracked object is not deleted even if it is not matched with an object detected in a subsequent frame.

### Claim 10

A unified matching system for multi-pedestrian tracking, comprising:
a controller comprising at least one processor configured to control a vehicle based on information on road environment images of a road on which the vehicle travels,
wherein the controller is configured to: divide the road environment images into frames; detect an object in an N-th frame; calculate a matching cost matrix between the detected object and tracked objects detected in a previous frame of the road environment images; match the detected object with one of the tracked objects by using the matching cost matrix; generate a matching threshold value based on a confidence score of the detected object; update the matched tracked object using the detected object when a matching cost between the matched tracked object and the detected object is smaller than the matching threshold value; and control the vehicle based on the updated tracked object.

### Claim 11 (depends on 10)

The unified matching system of claim 10, wherein the controller is connected to a braking apparatus configured to control a longitudinal driving of the vehicle and a steering apparatus configured to control a lateral driving of the vehicle, and
the controller is configured to control at least one of the braking apparatus and the steering apparatus based on the updated tracked object.

### Claim 12 (depends on 10)

The unified matching system of claim 10, wherein the information on the road environment images is collected by a sensor, and the sensor comprises at least one of a front camera and a rear camera.

### Claim 13 (depends on 10)

The unified matching system of claim 10, wherein the controller is configured to determine the confidence score of the detected object and generate the matching threshold value by multiplying the confidence score with a predetermined threshold constant value.

### Claim 14 (depends on 13)

The unified matching system of claim 13, wherein the controller is configured to generate a matching threshold matrix including matching threshold values, and
if the number of tracked objects is n and the number of detected objects is m, a dimension of the matching threshold matrix is (n, m).

### Claim 15 (depends on 14)

The unified matching system of claim 14, wherein the controller is configured to: predict expected positions of the tracked objects in the N-th frame by using a Kalman filter; and calculate the matching cost matrix based on Intersection over Union (IoU) values between bounding boxes of the tracked objects at the expected positions and a bounding box of the detected object.

### Claim 16 (depends on 15)

The unified matching system of claim 15, wherein the controller is configured to: calculate embedding vectors of the tracked objects by inputting regions of the bounding boxes of the tracked objects into a feature extractor; calculate an embedding vector of the detected object by inputting a region of the bounding box of the detected object into the feature extractor; and calculate the matching cost matrix based on cosine distances between the embedding vectors of the tracked objects and the embedding vector of the detected object.

### Claim 17 (depends on 16)

The unified matching system of claim 16, wherein the controller is configured to perform one-to-one matching between the detected object and the tracked objects by using a Hungarian algorithm.

### Claim 18 (depends on 17)

The unified matching system of claim 17, wherein the controller is configured to generate a new tracked object using the detected object when the confidence score of the detected object is greater than or equal to a predetermined threshold and the detected object is not matched with any one of the tracked objects.

### Claim 19 (depends on 18)

The unified matching system of claim 18, wherein the controller is configured to delete the new tracked object when the new tracked object is not matched with an object detected in a subsequent frame, determine the new tracked object as a confirmed tracked object when the new tracked object is subsequently matched for at least a predetermined number of consecutive frames, and the confirmed tracked object is not deleted even if the confirmed tracked object is not matched with an object detected in a subsequent frame.

### Claim 20

A non-transitory computer-readable recording medium that records a program for executing a unified matching method for multi-pedestrian tracking on a computer, the method comprising:
dividing road environment images collected from at least one sensor into frames and detecting an object in an N-th frame;
calculating a matching cost matrix between the detected object and tracked objects detected in a previous frame of the road environment images;
matching the detected object with one of the tracked objects using the matching cost matrix;
generating a matching threshold value based on a confidence score of the detected object;
updating the matched tracked object using the detected object when a matching cost between the matched tracked object and the detected object is smaller than the matching threshold value; and
controlling a vehicle based on the updated tracked object.
