High Precision Pallet Detection and Location Algorithm for AGV

Automated guided vehicle (AGV) is the primary solution for automated material handling, achieving precise identification and pickup without relying on external aids such as pallet limits, magnetic strips, or fixed paths is a challenging task. This paper proposes a high-precision pallet detection and positioning method based on a depth camera, which can accurately estimate the pallet pose in a factory environment and assist the AGV in completing the pallet picking task. The main contributions are as follows: 1. A pallet recognition method based on geometric feature matching of pallet legs is proposed, which extracts pallet legs through point cloud filtering, segmentation, and clustering, and combines pallet geometry feature detection to achieve a lightweight pallet recognition method. 2. A precise pose calculation method is designed, which quickly processes subsequent data based on the prior pose obtained from the first frame, selectively fuses features of multiple frames, weakens the fluctuations of single-frame data, and enriches the point cloud of pallet support pillars to improve the accuracy of pose calculation. The proposed method has been deployed and implemented on embedded devices , and has been verified and tested in an industrial environment with AGV. The results show that the pallet detection rate can reach 99%, and the positioning accuracy can reach 10mm and 0.5°, which can achieve accurate pallet pose estimation in practical application environments.