Convolutional Neural Network based on Channel-specific Convolution for High-Speed Object Recognition of 3D LiDAR

摘要

The present invention relates to a convolution-based convolutional neural network structure for each channel for high-speed object recognition of 3D LiDAR. An object using an out-of-channel convolutional network and a feature map generated in the out-of-channel convolutional network for extracting features between channels using the output for each channel of It includes a detection network that discovers the location and class of According to the present invention, there is an effect that data loss does not occur because raw data coming from the lidar is directly used in the high-speed object recognition process of the 3D lidar.