基于三维语义包围框的车辆事故检测方法研究

Research on Vehicle Accident Detection Method Based on 3D Semantic Bounding Box

为了研究在交通监控场景下的车辆空间运动状态,介绍了一种基于三维目标检测生成三维语义包围框的车辆事故检测方法。该方法从卷积神经网络中提取车辆运动姿态,预测出车辆的三维边界框。融合三维目标检测和二维到三维扩展卡尔曼滤波器,在视频序列中恢复6自由度的三维车辆姿态和跟踪轨迹,利用三维语义包围框建立事故检测评价指标。实验结果表明,所提出的联合二维和三维交通事故检测方法准确率可达81.75%。

In order to study the vehicle motion state in traffic monitoring scenarios,a vehicle accident detection method based on 3D object detection to generate 3D semantic bounding box is introduced.The method extracts the vehicle motion pose from the convolutional neural network and predicts the 3D bounding box of the vehicle.The 3D object detection and 2D are fused to 3D ex⁃tended Kalman filter to recover 3D vehicle pose with 6 degrees of freedom and trajectory for multiple vehicle tracking in the video se⁃quences.The 3D semantic bounding box is used to establish an accident detection evaluation index.The experiment results show that the accuracy of the proposed joint 2D and 3D traffic accident detection methods can reach 81.75%.