Skip to content

Latest commit

 

History

History
46 lines (26 loc) · 2.51 KB

README.md

File metadata and controls

46 lines (26 loc) · 2.51 KB

mmWave_radar_tracking

object tracking based on millimeter wave radar data

by dawnlh

Introduction

Object tracking based on millimeter wave radar data with Kalman Filter algorithm.

Flow chart for the whole project

single object tracking

Flow chart for the object tracking module

single object tracking

Result demo

single object tracking

single object tracking

Note

  • In multiple object tracking, when objects have overlapping, mistakes may occur. This problem can perhaps be solved by using a more robust detectionToTrackAssignment.m function. Specifically, we can take other statistic features into account when calculating the cost in detectionToTrackAssignment.m function. Currently, only position and speed are considerer.
  • The performance of the implemented algorithm is very dependent on parameter tuning, especially the parameters of DBSCAN and the tracking module (like parameters in detectionToTrackAssignment.m and updateTrackStates.m).
  • Tips to improve the performance
    • A more sophisticated denoising algorithm. Current point_cloud_denoise.m only removes the static points and out-of-range points, the "real" noise is not filtered. A better denoising algorithm may consider the spatial-temporal information between adjacent frames and adjacent regions.
    • A more sophisticated cluster algorithm, which can automatically figure out different objects in tough cases like overlapping.
    • A more sophisticated tracking algorithm and strategy, which take more information into consideration and realize better performance.
    • Optimize the code to lower the computation cost and speed up the running speed.

Reference