Improving 3D Object detection using contextual information & 2D features

Abstract

3D object detection is becoming an active research topic in both computer vision. Compared to 2D object detection in RGB images, predicting 3D

bounding boxes in real world environments captured by point clouds is more essential for many tasks, such as indoor robot navigation, robot grasping, etc. However, the unstructured data in point clouds makes the detection more challenging than in 2D. In particular, the popular convolutional neural networks (CNNs), which are highly successful in 2D object detection, are difficult to be applied to point clouds directly. In this paper, we present a new model (CIMVNet) for object recognition within the 3D environment based on enhancing the 3D image information with a set of features extracted from the 2D color images, in addition to taking advantage of the contextual information of the scene and the correlations of its various components to improve the accuracy of the 3D features. The developed model proved that taking advantage of more than one source of data within the 3D image improves the accuracy of classification. It has achieved an accuracy improvement compared to the state- of-the-art researches, which amounted to 3.04% over the highest previous reference research