A Pallet Recognition Method Based on Adaptive Color Fast Point Feature Histogram

Abstract

Pallet recognition is one of the critical technologies of cargo handling for unmanned industrial vehicles. A pallet recognition method based on adaptive color fast point feature histogram (ACFPFH) is proposed to solve the problems of current recognition methods such as low efficiency, time-consuming, poor robustness and random parameter selection. The Kinect V2 sensor is used to collect the point cloud data which represents the whole scene including pallet. Next, outliers are removed and the optimal neighborhood radius of each point is obtained based on the minimum criterion of neighborhood feature entropy function. Then, the key points are extracted from scene point clouds. The ACFPFH consisting of color feature and adaptive geometric feature is applied for similarity matching between the template and scene point clouds. Finally, wrong feature correspondences are rejected and the pallet in the scene point cloud is recognized. A comparison of the fast point feature histogram with the fixed neighbor radius of 0.012 m shows that the pallet recognition precision and efficiency of the method based on ACFPFH is improved by 83.74% and 35.55% respectively, which verifies the superiority of the proposed method.

Keywords

• fast point feature histogram
• feature extraction
• adaptive optimal neighborhood
• pallet recognition