Abstract: The local feature description of point clouds is a fundamental but challenging task in the fields of image and vision. The existing descriptors are mainly real-valued features, which have problems such as lengthy dimensions, low matching and storage efficiency. In view of this, this paper proposes a binary encoding in invariant voxel space descriptor. Firstly, while maintaining rotational invariance, in order to fully exploit the geometric information of point clouds in Euclidean space and attribute space, we project the local point cloud into a local reference frame and invariant distance attribute space. Then, we voxelized the point clouds in two types of spaces and proposed binary voxel attribute testing methods for single source attributes and cross source attributes, fully decoding the complementary geometric information of the two types of spaces. Finally, an information entropy guided bit selection method is proposed to remove interfering bits while retaining rich information bits, forming the final descriptor. The experimental results of B3R, RESSO-indoor, and RESSO-outdoor datasets show that BEIVS performs the best in all datasets, surpassing existing real-valued descriptors, with recall and accuracy curve AUC(the area under the recall-precision curve) values of 0.999, 0.047, and 0.193. The effectiveness of the method is verified.