KR-20260067715-A - Object tracking system for tracking occluded objects using feature points

Abstract

The present invention relates to an object tracking system that tracks an occluded object using feature points. According to the present invention, the shape of an object scanned by a LiDAR is periodically updated so that it can accurately determine whether it is the same object even when an occlusion phenomenon occurs.

Inventors

• 박만복

Assignees

• 국립한국교통대학교산학협력단

Dates

Publication Date

20260513

Application Date

20241106

Claims (3)

1. An object recognition unit that recognizes objects from scan data acquired through LiDAR; A feature point extraction unit that extracts feature points from an object recognized by the object recognition unit above; A contour information generation unit that generates contour information of the corresponding object through feature points extracted by the above feature point extraction unit; and An object tracking system characterized by including: an object tracking unit that assigns an ID to an object recognized by the object recognition unit, updates contour information generated by the contour information generation unit by matching it to the ID of the object, and determines whether to maintain the same ID or assign a new ID based on whether the contour information of the object recognized in the current scan data matches the contour information of the object recognized in the previous scan data.
2. In paragraph 1, The above feature point extraction unit extracts the start point and end point as feature points from the scan data of the recognized object, and An object tracking system characterized by the above-described contour information generation unit generating a virtual baseline connecting the start point and the end point, extracting the scan data furthest from the baseline as an intermediate point, and then generating a first contour line connecting the start point and the intermediate point and a second contour line connecting the intermediate point and the end point as contour information.
3. In paragraph 2, The object tracking system is characterized by the above object tracking unit comparing the baseline, first contour, and second contour, which are contour information of an object recognized in the scan data of the current round, with the baseline, first contour, and second contour, which are contour information of an object recognized in the scan data of the previous round, and maintaining the same ID if it is confirmed that one or more match, and assigning a new ID if the contour information does not all match.

Description

Object tracking system for tracking occluded objects using feature points The present invention relates to an object tracking system, and more specifically, to a technology that utilizes feature points of an object recognized by LiDAR to maintain the status of the same object even when the object is obscured by another object. In the fields of smart cars and autonomous vehicles, object recognition is a fundamental yet essential technical element for safety. Of course, there are also cases where infrastructure sensors, rather than moving vehicles, detect and track surrounding objects. Environmental sensors such as cameras or LiDAR are used for object recognition. However, analyzing video captured by cameras only confirms the presence of an object and has the problem of not being able to accurately determine its relative distance. On the other hand, using LiDAR allows for the accurate identification of not only the presence of an object but also its shape and relative distance. However, even in the case of object recognition using LiDAR, there is a problem in that it cannot accurately determine whether multiple objects are the same when they overlap. That is, as illustrated in (a) of FIG. 1, when the surroundings are scanned by the lidar (10), if the first object (OBJ1) and the second object (OBJ2) are separated from each other in the scan area, each object can be correctly recognized and assigned a unique ID. However, as shown in (b) of FIG. 1, when the first object (OBJ1) moves and is obscured behind the second object (OBJ2), the scan data of the lidar (10) can recognize that the second object (OBJ2) and the object located behind it are different objects, but the object scanned from the left of the second object (OBJ2) (OBJ1) and the object scanned from the right (OBJ3) are determined to be different objects. That is, because continuity is lost in the scan data, although they are actually the same object, they can be recognized as being separated into the first object (OBJ1) and the third object (OBJ3). Since accurately distinguishing and tracking objects is closely related to safety, recognizing identical objects as separate entities can lead to fatal tracking errors. Meanwhile, prior art for tracking objects using LiDAR includes Korean Patent Publication No. 10-2022-0126472 (September 16, 2022, 'Method and apparatus for tracking objects using a LiDAR sensor, vehicle including the apparatus, and recording medium recording a program for executing the method'). FIG. 1 is a diagram illustrating an example of recognizing the same object as different objects through a conventional LiDAR. FIG. 2 is a block diagram illustrating an object tracking system according to an embodiment of the present invention. FIG. 3 is a flowchart for explaining the process of tracking an object through the object tracking system illustrated in FIG. 2. FIG. 4 is a diagram illustrating an example of generating contour information of an object using feature points in the object tracking system illustrated in FIG. 2. FIG. 5 is a diagram illustrating an example in which an object is determined to be the same object even when an occlusion phenomenon occurs when an object is recognized through the object tracking system illustrated in FIG. 2. FIG. 6 is a diagram illustrating an example of distinguishing different objects when an occlusion phenomenon occurs when an object is recognized through the object tracking system illustrated in FIG. 2. Preferred embodiments of the present invention will be described below with reference to the accompanying drawings. However, some components unrelated to the gist of the invention may be omitted or compressed; nevertheless, omitted components are not necessarily unnecessary for the present invention and may be combined and used by those skilled in the art to which the present invention pertains. FIG. 2 is a block diagram illustrating an object tracking system according to an embodiment of the present invention. As shown in FIG. 2, the object tracking system (20) according to an embodiment of the present invention includes an object recognition unit (21), a feature point extraction unit (22), a contour information generation unit (23), and an object tracking unit (24). This object tracking system (20) is configured to recognize surrounding objects through scan data from the LiDAR (10), assign a unique ID to each object, and then track it. Depending on the implementation, the object tracking system (20) and the LiDAR (10) may be installed in an autonomous vehicle or in an infrastructure sensor. Of course, the autonomous vehicle or infrastructure sensor may include other system configurations for autonomous driving or surveillance, but FIG. 2 only illustrates the configurations necessary for object tracking. The object recognition unit (21) is provided to recognize and separate each object from the scan data (BP) of the lidar (10). The feature point extraction unit (22) is provided to extract feature poin