A portable three-dimensional LIDARbased system for long-term and widearea people behavior measurement

https://journals.sagepub.com/doi/pdf/10.1177/1729881419841532

1. Introduction

2. Related work

3. System overview

4. Offline environmental mapping

5. Online people behavior measurement

5.1 Sensor localization

5.2 People detection and tracking

먼저 foreground points 추출을 위해서 배경을 제거 하였다. We first remove the background points from an observed point cloud to extract the foreground points.

이후, 복셀사이즈 0.5m로 occupancy grid map를 생성 하였다. Then, we create an occupancy grid map with a certain voxel size (e.g.0.5 m) from the environmental map.

센서 자세 추정을 통해 지도 좌표계로 입력된 포인트 클라우드를 변경 하였다. 이후 environmental map에 포함된 포인트를 배경으로 간주 하고 제거 하였다. The input point cloud is transformed into the map coordinate according to the sensor pose estimated by UKF, and then each point at a voxel containing environmental map points is removed as the background.

후보 사람 영역 클러스터링을 위해 유클리드 군집화 기법을 적용 하였다. The Euclidean clustering is then applied to the foreground points to detect human candidate clusters.

하지만, 근접된 사람들은 하나의 클러스터로 구분 되는 문제가 있다. However, in case persons are close together, their clusters may be wrongly merged and are detected as a single cluster.

문제 해결을 위해 Haselich’s split merge clustering알고리즘을 적용 하였다. To deal with this problem, we employ Haselich’s split merge clustering algorithm.깃북 정리

Confidence-Based Pedestrian Tracking in Unstructured Environments Using 3D Laser Distance Measurements

Haselich’s split merge clustering알고리즘은 클러스터를 서브-클러스터로 특정 쓰레쉬홀드까지 dp-means를 이용하여 작게 나눈다. The algorithm first divides a cluster into subclusters until each cluster gets smaller than a threshold (e.g. 0.45m) by using dp-means so that every cluster does not have points of different persons.

만약 서브-클러스터 사이에 간격이 없다면 하나의 사람으로 간주되고 머징된다. Then, if there is no gap between those subclusters, the clusters are considered to belong to a single person and remerged into one cluster.

Figure 10 shows an example of the detection results.

위 방식을 통해 근접한 두 사람도 잘 나뉘게 된다. The person clusters are correctly separated even when they are very close together thanks to the split and the remerge process

탐지된 클러스터는 사람이 아닌것도 포함될수 있다. The detected clusters may contain nonhuman clusters (i.e. false positives).

사람인지 아닌지 분류 작업을 Kidono의 알고리듬을 이용하여 진행 한다깃북정리 . To eliminate nonhuman clusters among detected clusters, we judge whether a cluster is a human or not by using a human classifier trained with slice features by Kidono et al. and Schapire and Singer.

Pedestrian Recognition Using High-definition LIDAR

사람이 지표면위에 걸어 다닌다는 가정 하에 높이 정보 상관 없이 xy로 추적을 실시 한다. Assuming that persons walk on the ground plane, we track persons on the XY plane without the height.

추적을 위해서 아래 방법들을 사용 하였다. We employ the combination of Kalman filter with the constant velocity model and global nearest neighbor data association to track persons.

• Kalman filter
• constant velocity mode
• global nearest neighbor data association

제안된 기법은 잘 동작 한다. The tracking scheme works well as long as the tracked persons are visible from the sensor and are correctly detected.

hdl_people_tracking

hdl_people_tracking is a ROS package for real-time people tracking using a 3D LIDAR.

• 클러스터링 : It first performs Haselich's clustering technique to detect human candidate clusters,
  • Confidence-Based Pedestrian Tracking in Unstructured Environments Using 3D Laser Distance Measurements
  • 깃북 정리
• 분류 : and then applies Kidono's person classifier to eliminate false detections.
  • Pedestrian Recognition Using High-definition LIDAR
  • 깃북정리
• 추적 : The detected clusters are tracked by using Kalman filter with a contant velocity model.