Fast Dynamic Vehicle Detection in Road Scenarios Based on Pose Estimation with Convex-Hull Model.

Kaiqi Liu,Jianqiang Wang

doi:10.3390/s19143136

Abstract

Dynamic vehicle detection is of great significance for the safety of autonomous vehicles and the formulation of subsequent driving strategies. A pose-estimation algorithm, namely, the pose estimation with convex-hull model (PE-CHM), is proposed in this paper, and introduced in the dynamic vehicle detection system. In PE-CHM, the convex hull of the object’s point-clouds is first extracted and the most fitted bounding box is determined by a multifactor objective function. Next, the center position of the target is inferred according to the location and direction of the target. With the obtained bounding box and the position inference, the pose of the target is determined, which reduces the interference of the missing contour on pose estimation. Finally, three experiments were performed to validate the performance of the proposed PE-CHM method. Compared with several typical model-based methods, PE-CHM can implement dynamic vehicle detection faster, which reduces the amount of calculation on the basis of ensuring detection efficiency.

Highlights

There are generally static and moving obstacles in traffic environments
Compared with several typical model-based methods, pose estimation with convex-hull model (PE-Convex-hull model (CHM)) can implement dynamic vehicle detection faster, which reduces the amount of calculation on the basis of ensuring detection efficiency
The distance information is of great significance in the formulation of driving strategies for autonomous vehicles, which can be obtained with a Lidar sensor

Summary

Introduction

There are generally static and moving obstacles in traffic environments. For autonomous vehicles in road scenario, since moving objects have their own movement state and motion direction, their motion randomness affects the behavior decision of an autonomous vehicle, which poses a greater threat to traffic safety. Camera and Lidar are commonly used sensors. A dynamic object detection method in monitoring scenario with fixed cameras has been proposed [7]. The method models each pixel as mixture of Gaussians and can deal with slow lighting changes and multimodal distributions caused by shadows, specularities, swaying branches and other troublesome features of the real world. This kind of method is not suitable to detect the moving objects for the mobile platform. The distance information is of great significance in the formulation of driving strategies for autonomous vehicles, which can be obtained with a Lidar sensor

Objectives

Results

Conclusion