Deep learning with 3D data such as reconstructed point clouds and CAD models has received great research interests recently. However, the capability of using point clouds with convolutional neural network has been so far not fully explored. In this paper, we present a convolutional neural network for semantic segmentation and object recognition with 3D point clouds. At the core of our network is pointwise convolution, a new convolution operator that can be applied at each point of a point cloud. Our fully convolutional network design, while being surprisingly simple to implement, can yield competitive accuracy in both semantic segmentation and object recognition task.

No abstract

No abstract

No abstract

Existing deep learning-based obstacle detection systems are often designed and implemented based on raw input feature. These systems obtain high accuracy under normal driving conditions. But they fail to operate under difficult driving conditions, which are different from their training. Recently, an unsupervised auto-encoder has been successfully applied to produce robust input features for a stereo matching system under difficult driving conditions. Therefore, this paper investigates an auto-encoder feature to improve the performance of existing vehicle detections under adverse weather conditions. Experimental results show that the proposed method obtained better result than existing state-of-the-art object detection methods in term of accuracy. Index Terms-Vehicle detection, auto-encoder, deep learning, and local binary pattern. Vinh Dinh Nguyen received the B.Sc. (magna cum laude