# Extended Classification Course Improves Road Intersection Detection from Low-Frequency GPS Trajectory Data

^{1}

^{2}

^{3}

^{4}

^{*}

## Abstract

**:**

## 1. Introduction

- (1)
- We introduce a stay-point detection algorithm in the pre-processing course to remove a large proportion of false turn-points not indicating road intersections, which few methods have considered before;
- (2)
- We propose a turn-point compensation algorithm, which utilizes geometric and hot area features of road intersection. This algorithm retrieves actual turning positions from low-frequency trajectories using a turning angle assumption determined by the density map of all GPS points. An indicator to evaluate the quality of a selected turn-point is presented, proving that the concentration of turn-points is significantly improved by our pre-processing and position compensation processes;
- (3)
- We extend the point clustering-based road intersection detection framework to include a post-classification course. The clustering algorithm first yields a recall-focused detection result, following which the classifier greatly increases the detection precision by utilizing the geometric features of road intersections, with a small cost in recall rate due to misclassification. The extended classification course makes the overall performance of our method better than that of existing precision-focused methods, and it is easy to implement co-operatively with any recall-focused detection algorithm.

## 2. Related Work

## 3. Road Intersection Detection from Low-Frequency Trajectory Data

#### 3.1. Trajectory Segmentation Based on Stay-Point Detection

#### 3.2. Clustering Turn-Points after Position Compensation

#### 3.2.1. Turn-Point Compensation Based on Turning Angle Assumption

#### 3.2.2. Clustering Algorithm Based on Delaunay Triangulation

#### 3.3. Road Intersection Classification Based on Thinning Algorithm

#### 3.3.1. Road Centerline Extraction from Density Map

#### 3.3.2. Classifying Road Intersections Based on Road Centerlines

## 4. Experimental Results

#### 4.1. Datasets and Detection Evaluation

**Chicago Dataset**. In our experiment, the Chicago dataset, which has a sample rate of 3.6 s, was used to represent high-quality data. It is a rather small dataset, covering an area of 3.0 $\times $ 1.9 km and containing 889 traces along with 118,360 GPS points.

**Shenzhen Dataset**. The Shenzhen dataset, which has a sample rate of 26.1 s, represents low-quality data. It contains over 75 million GPS points from 14,716 taxis, recorded over the course of one day [25]. In our experiments, only a subset which covers an area of 2.0 $\times $ 2.0 km was used. The longitude and latitude of this area’s center point is (114.08°,22.55°). The subset contains 19,906 traces and 319,601 GPS points.

**Detection Precision and Recall**. The road intersection detection result was evaluated by computing the precision, recall, and ${\mathrm{F}}_{\mathrm{score}}$. A set of distance thresholds were used in matching detected intersections with the ground truths offered by OpenStreetMap [21]. Given a distance threshold $d$, if a detected intersection’s center position was within $d$ of a ground truth’s center position, we counted it as a matched intersection. The precision, recall, and ${\mathrm{F}}_{\mathrm{score}}$ are defined as follows:

#### 4.2. Evaluating Detection of Road Intersections

#### 4.2.1. Chicago Dataset

#### 4.2.2. Shenzhen Dataset

#### 4.2.3. Effectiveness of Key Algorithms

#### 4.3. Parameter Analysis

#### 4.3.1. Turn-Point Selection

#### 4.3.2. Clustering

#### 4.3.3. Sensitivity to Data Volume

## 5. Conclusions

## Author Contributions

## Funding

## Acknowledgments

## Conflicts of Interest

## References

- TomTom Map Update Service. Available online: www.tomtom.com/en_gb/maps/map-update-service/ (accessed on 22 July 2011).
- Wang, W.; Yang, N.; Zhang, Y.; Wang, F.; Cao, T.; Eklund, P. A review of road extraction from remote sensing images. J. Traffic Transp. Eng. Engl. Ed.
**2016**, 3, 271–282. [Google Scholar] [CrossRef][Green Version] - Mena, J. State of the art on automatic road extraction for GIS update: A novel classification. Pattern Recognit. Lett.
**2003**, 24, 3037–3058. [Google Scholar] [CrossRef] - Guan, H.; Li, J.; Cao, S.; Yu, Y. Use of mobile LiDAR in road information inventory: A review. Int. J. Image Data Fusion
**2016**, 7, 219–242. [Google Scholar] [CrossRef] - Goodchild, M.F. Citizens as voluntary sensors: Spatial data infrastructure in the world of web 2.0. Int. J. Spatial Data Infrastruct. Res.
**2007**, 2, 24–32. [Google Scholar] - Ahmed, M.; Karagiorgou, S.; Pfoser, D.; Wenk, C. A comparison and evaluation of map construction algorithms using vehicle tracking data. GeoInformatica
**2014**, 19, 601–632. [Google Scholar] [CrossRef][Green Version] - Available online: https://www.cs.rutgers.edu/~dz220/data.html (accessed on 22 October 2018).
- Edelkamp, S.; Schrödl, S. Route Planning and Map Inference with Global Positioning Traces. Comput. Vis.
**2003**, 2598, 128–151. [Google Scholar] - Peng, C.; Jin, X.; Wong, K.C.; Shi, M.; Liò, P. Collective human mobility pattern from taxi trips in urban area. PLoS ONE
**2012**, 7, e34487. [Google Scholar] - Deng, M.; Huang, J.; Zhang, Y.; Liu, H.; Tang, L.; Tang, J.; Yang, X. Generating urban road intersection models from low-frequency GPS trajectory data. Int. J. Geogr. Inf. Sci.
**2018**, 32, 1–25. [Google Scholar] [CrossRef] - Chen, C.; Lu, C.; Huang, Q.; Yang, Q.; Gunopulos, D.; Guibas, L. City-Scale Map Creation and Updating using GPS Collections. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining–KDD ’16, San Francisco, CA, USA, 13–17 August 2016; pp. 1465–1474. [Google Scholar]
- Zhang, C.; Xiang, L.; Li, S.; Wang, D. An Intersection-First Approach for Road Network Generation from Crowd-Sourced Vehicle Trajectories. ISPRS Int. J. Geo Inf.
**2019**, 8, 473. [Google Scholar] [CrossRef][Green Version] - Karagiorgou, S.; Pfoser, D. On vehicle tracking data-based road network generation. In Proceedings of the 20th International Conference on Intelligent User Interfaces Companion–IUI Companion ’15, Lyon, France, 16–20 April 2012; p. 89. [Google Scholar]
- Wu, J.; Zhu, Y.; Ku, T.; Wang, L. Detecting Road Intersections from Coarse-gained GPS Traces Based on Clustering. J. Comput.
**2013**, 8, 2959–2965. [Google Scholar] [CrossRef] - Schroedl, S.; Wagstaff, K.L.; Rogers, S.; Langley, P.; Wilson, C. Mining GPS Traces for Map Refinement. Data Min. Knowl. Discov.
**2004**, 9, 59–87. [Google Scholar] [CrossRef] - Ahmed, M.; Wenk, C. Constructing Street Networks from GPS Trajectories. In Algorithms–ESA2012; Springer: Berlin/Heidelberg, Germany, 2012; pp. 60–71. [Google Scholar]
- Fathi, A.; Krumm, J. Detecting Road Intersections from GPS Traces. In Geographic Information Science; Springer: Berlin/Heidelberg, Germany, 2010; pp. 56–69. [Google Scholar]
- Xie, X.; Liao, W.; Aghajan, H.; Veelaert, P.; Philips, W. Detecting Road Intersections from GPS Traces Using Longest Common Subsequence Algorithm. ISPRS Int. J. Geo Inf.
**2016**, 6, 1. [Google Scholar] [CrossRef][Green Version] - Zhang, L.; Thiemann, F.; Sester, M. Integration of GPS traces with road map. In Proceedings of the Second International Workshop on Data-Aware Distributed Computing–DADC ’09, Munich, Germany, 9 June 2010; pp. 17–22. [Google Scholar]
- Li, Q.; Zheng, Y.; Xie, X.; Chen, Y.; Liu, W.; Ma, W.-Y. Mining user similarity based on location history. In Proceedings of the 16th ACM SIGSPATIAL International Conference, Irvine, CA, USA, 5–7 November 2008; p. 1. [Google Scholar]
- OpenStreetMap. Available online: http://www.openstreetmap.org/ (accessed on 22 October 2018).
- Biagioni, J.; Eriksson, J. Map inference in the face of noise and disparity. In Proceedings of the 20th International Conference on Intelligent User Interfaces Companion–IUI Companion ’15, Lyon, France, 16–20 April 2012; pp. 79–88. [Google Scholar]
- Yang, X.; Cui, W. A Novel Spatial Clustering Algorithm Based on Delaunay Triangulation. J. Softw. Eng. Appl.
**2010**, 3, 141–149. [Google Scholar] [CrossRef][Green Version] - Zhang, T.Y.; Suen, C.Y. A fast parallel algorithm for thinning digital patterns. Commun. ACM
**1984**, 27, 236–239. [Google Scholar] [CrossRef] - Zhang, D.; Zhao, J.; Zhang, F.; He, T. UrbanCPS: A cyber-physical system based on multi-source big infrastructure data for heterogeneous model integration. In Proceedings of the ACM/IEEE 6th International Conference on Cyber-Physical Systems (ICCPS’15), Seattle, WA, USA, 14–16 April 2015; pp. 238–247. [Google Scholar]

**Figure 2.**Kernel density estimation of stay-points. The red circles are regions with road intersection.

**Figure 4.**Turn-point compensation process: (

**a**) compensation position by Wu’s method; (

**b**) series of candidate positions by our method; (

**c**) density map helps to determine the optimal turn-point.

**Figure 8.**Extract road centerlines by thinning an image slice; (

**a**) from a skewed T-shaped intersection. (

**b**) from a road segment; and (

**c**) result with a spurious road due to GPS noise.

**Figure 11.**Detection precision, recall, and ${\mathrm{F}}_{\mathrm{score}}$. (

**a**) comparing detection result with non-classified; (

**b**) comparing detection result with non-compensated.

**Figure 13.**Detection ${\mathrm{F}}_{\mathrm{score}}$ with different cluster algorithm settings; (

**a**) using different cluster radius thresholds; (

**b**) using different cluster-points thresholds.

Attribute | Symbol | Description | Unit |
---|---|---|---|

Latitude | $la{t}_{i}$ | Latitude | degree |

Longitude | $lo{n}_{i}$ | Longitude | degree |

Timestamp | ${t}_{i}$ | UTC time when the GPS measurement was taken | s |

Speed | ${v}_{i}$ | Estimated using current point and its predecessor | m/s |

Heading | ${h}_{i}$ | Estimated using current point and its predecessor | degree |

Dataset | Arbitrary Points | Turn-Points |
---|---|---|

Chicago campus bus | ${T}_{50}=33.84\%$ | ${T}_{50}=90.30\%$ |

Shenzhen taxi | ${T}_{50}=39.88\%$ | ${T}_{50}=50.14\%$ |

Classification Accuracy = 80% | Classification Accuracy = 90% | ||
---|---|---|---|

$\left(\mathrm{P},\mathrm{R}\right)$ before classification | $\left({\mathrm{P}}^{*},{\mathrm{R}}^{*}\right)$ after classification | $\left(\mathrm{P},\mathrm{R}\right)$ before classification | $\left({\mathrm{P}}^{*},{\mathrm{R}}^{*}\right)$ after classification |

(90%, 60%) | (97.3%, 48%) | (90%, 60%) | (98.7%, 54%) |

(80%, 70%) | (94.1%, 56%) | (80%, 70%) | (97.2%, 63%) |

(70%, 80%) | (90.3%, 64%) | (70%, 80%) | (95.4%, 72%) |

(60%, 90%) | (85.7%, 72%) | (60%, 90%) | (93.1%, 81%) |

Match Threshold | 5 m | 10 m | 20 m | 40 m |
---|---|---|---|---|

8991 turn-points being selected by Karagiorgou’s criterion | ||||

Precision | 12.24% | 38.78% | 77.55% | 89.80% |

Recall | 12.00% | 38.00% | 76.00% | 88.00% |

3951 turn-points being selected by our criterion | ||||

Precision | 13.51% | 62.16% | 89.19% | 100.00% |

Recall | 10.00% | 46.00% | 66.00% | 74.00% |

Match Threshold | Precision | Recall | ${F}_{score}$ | |
---|---|---|---|---|

Proposed | 40 m | 51/58 = 87.93% | 50/66 = 75.76% | 0.82 |

Karagiorgou | 45/76 = 59.21% | 39/66 = 59.09% | 0.59 |

Truth | Classified as Intersection | Classified as Road Segment |
---|---|---|

Intersection | 0.78 | 0.22 |

Road Segment | 0.09 | 0.91 |

Angle Thres | Speed Thres | Match Thres | Precision | Recall | |
---|---|---|---|---|---|

Proposed | ≥45° | $\le $10 m/s | 87.93% | 75.76% | |

Karagiorgou | ≥15° | $\le $11.1 m/s | 40 m | 85.00% | 77.27% |

Wu | ≥45° | N/A | 85.96% | 74.24% |

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Chen, B.; Ding, C.; Ren, W.; Xu, G. Extended Classification Course Improves Road Intersection Detection from Low-Frequency GPS Trajectory Data. *ISPRS Int. J. Geo-Inf.* **2020**, *9*, 181.
https://doi.org/10.3390/ijgi9030181

**AMA Style**

Chen B, Ding C, Ren W, Xu G. Extended Classification Course Improves Road Intersection Detection from Low-Frequency GPS Trajectory Data. *ISPRS International Journal of Geo-Information*. 2020; 9(3):181.
https://doi.org/10.3390/ijgi9030181

**Chicago/Turabian Style**

Chen, Banqiao, Chibiao Ding, Wenjuan Ren, and Guangluan Xu. 2020. "Extended Classification Course Improves Road Intersection Detection from Low-Frequency GPS Trajectory Data" *ISPRS International Journal of Geo-Information* 9, no. 3: 181.
https://doi.org/10.3390/ijgi9030181