Abstract
Three-dimensional (3D) object detection is an important research in 3D computer vision with significant applications in many fields, such as automatic driving, robotics, and human–computer interaction. However, the low precision is an urgent problem in the field of 3D object detection. To solve it, we present a framework for 3D object detection in point cloud. To be specific, a designed Backbone Network is used to make fusion of low-level features and high-level features, which makes full use of various information advantages. Moreover, the two-dimensional (2D) Generalized Intersection over Union is extended to 3D use as part of the loss function in our framework. Empirical experiments of Car, Cyclist, and Pedestrian detection have been conducted respectively on the KITTI benchmark. Experimental results with average precision (AP) have shown the effectiveness of the proposed network.
Highlights
The task of object detection is to find the objects of interest in a given scene and determine their category and location
Sensors 2019, 19, 4093 problems, we propose our solutions: 1. Only point cloud is used for 3D object detection to reduce the time cost, 2
Motivated by [20], we propose a 3D Generalized Intersection over Union (GIoU) loss function for 3D object detection, detection, which contributes to align 3D predicted and ground truth bounding boxes
Summary
The task of object detection is to find the objects of interest in a given scene and determine their category and location. Due to the loss of sophisticated spatial structure information in the process of projecting a Currently, deep learning-based 3D object detection in point cloud algorithms has a main challenge: the low detection precision. To solve this problem, some technologies [5,15] use a 2D detection algorithm in an image to locate the object, use bounding box regression to achieve 3D object detection. According to the detection results of a KITTI data set [16], they have achieved good results thanks to the accurate 2D detection in images These methods have two problems: they are highly dependent on 2D object detection technology and have an expensive time cost.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have