Aiming at the inaccurate detection of cone buckets by Euclidean clustering algorithm in the race track environment, a cone bucket detection method based on improved Euclidean clustering algorithm is proposed. Firstly, the point clouds are collected through the robot operating system (ROS); then the point clouds are preprocessed to find the region of interest (ROI) and then the ground and cone bucket point clouds are separated using a random sampling algorithm; then the distances and thresholds are modeled; then a region partitioning method is designed for the track environment to improve Euclidean clustering algorithm, and the cone bucket point clouds are segmented using dynamic threshold clustering; finally the algorithm is validated through the Matlab platform. The accuracy of clustering segmentation reached 93.98% and 99% in real-world tests under two track environments, respectively. The test results show that the proposed method can accurately detect the cone bucket in the track.