Pedestrian Group Detection with K-Means and DBSCAN Clustering Methods
Mingzuoyang Chen, Shadi Banitaan, Mina Maleki, Yichun Li · 2022
The development of autonomous vehicles has made real-time pedestrian detection and tracking an important research area for protecting human lives and improving society. A key challenge in this area is to improve pedestrian detection accuracy while reducing the processing time for tracking. This challenge can be addressed by detecting and tracking pedestrian groups since pedestrians move in groups. This paper proposes a method of detecting pedestrian groups using unsupervised machine learning methods. After detecting the pedestrians, K-Means and DBSCAN clustering methods were used to detect pedestrian groups based on the coordinates of pedestrians’ bounding boxes. Simulation results of the MOT17 dataset indicate that both clustering techniques could reduce detection and tracking processing times. However, K-Means clustering is more effective than DBSCAN clustering in detecting pedestrian groups as measured by the Silhouette Coefficient score and Adjusted Rand Index (ARI).