Apple Researchers Tout New Object-Detection Tech

Apple Inc. says it is developing a new lidar-based object-detection technology that promises improved performance over current lidar and camera sensors in self-driving vehicles.

Called VoxelNet, the technology is described as a generic 3D detection network that unifies feature extraction and bounding box prediction into a “single stage, end-to-end trainable deep network.” Apple researchers Yin Zhou and Oncel Tuzel detailed the technology in a paper published this month in the independent online journal arXiv.

VoxelNet divides a “point cloud” into equally spaced 3D voxels (values on a three-dimensional grid) to convert a group of points within each voxel into a unified representation. The grouping is encoded as a descriptive volumetric representation, which is then connected to a region-proposal network to generate detections.

Computer simulations have shown that VoxelNet outperforms the latest lidar-based 3D detection methods by a “large margin,” the researchers claim. The network’s discriminative representation of objects with various geometries also allows effective 3D detection of pedestrians and cyclists using lidar alone without the other sensors currently needed, according to the paper.

Late last year Apple acknowledged it was “investing heavily” in automated systems, but it has remained extremely secretive about the scope of its automotive program. This summer CEO Tim Cook said the company is focusing on autonomous systems, which he describes as "the mother of all AI projects."