Evaluation of Cluster Algorithms for Radar-Based Object Recognition in Autonomous and Assisted Driving

Area of Science:

• Automotive Engineering
• Computer Vision
• Sensor Fusion

Background:

• Perception systems are crucial for assisted driving and autonomous vehicles.
• Sensors like RADAR, cameras, and LIDAR provide environmental data for navigation.
• Radar technology offers robust object detection, especially in adverse weather conditions, using point cloud data.

Purpose of the Study:

• To evaluate the suitability of various clustering algorithms for automotive radar systems.
• To identify which algorithms are most effective for object identification, investigation, and tracking.
• To compare the performance of different clustering methods in the context of radar-based perception.

Main Methods:

• A comprehensive review of current clustering algorithms was conducted.
• The mathematical underpinnings of each algorithm were analyzed.
• Performance indicators were used to assess suitability and efficiency for radar applications.

Main Results:

• K-Means, Mean Shift, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) were found suitable for automotive radar.
• DBSCAN exhibited superior performance compared to other evaluated algorithms.
• The type of radar sensor significantly influences the effectiveness of object recognition methods.

Conclusions:

• DBSCAN is a highly effective algorithm for object recognition in automotive radar systems.
• The selection of appropriate clustering algorithms is vital for robust autonomous driving perception.
• Future research should consider the interplay between radar hardware and software algorithms.