基于双重注意力机制、轨迹预测与无人机遥感技术的人车流量检测方法

Pedestrian and Vehicle Flow Detection Method Based on Dual Attention Mechanism, Trajectory Prediction, and Drone Remote Sensing

人、车流量的精确检测统计对于公共安全和资源管理有重要意义。针对现有检测统计方法设备投入大、维护成本高、检测区域受限、易受环境因素影响等问题,提出一种基于双重注意力机制、轨迹预测与无人机遥感技术的人车流量检测方法。该方法在RefineDet网络的基础上加以改进,通过引入双重注意力机制、使用特征融合RNN替换TCB模块等方式增强对小目标的识别能力。同时通过自适应卡尔曼滤波预测轨迹,实现了对运动目标的实时跟踪,避免了因目标距离过远、行动方式特殊导致的错误统计。自建数据集训练网络模型,并在不同场景下进行测试。实验结果表明,所提改进模型相较对照模型的MOTA提高了2.3%~3.3%,在精度接近最新模型的同时检测速度更快,基本达到实时性需求;在多种场景下的综合评价指标F值均能达到95%以上,在多种实际场景检测中的误检率均低于0.3%。

Accurate detection and statistics of human and vehicle traffic flow are of great significance for public safety and resource manage-ment.A pedestrian and vehicle flow detection method based on dual attention mechanism,trajectory prediction,and drone remote sensing technology is proposed to address the issues of high equipment investment,high maintenance costs,limited detection areas,and susceptibility to environmental factors in existing detection and statistical methods.This method is improved on the basis of the RefineDet network by intro-ducing a dual attention mechanism and replacing the TCB module with feature fusion RNN to enhance the recognition ability of small targets.At the same time,adaptive Kalman filtering is used to predict trajectories,achieving real-time tracking of moving targets,avoiding erroneous statistics caused by target distance and special action modes.Train network models with self built datasets and conduct tests in different scenar-ios.The experimental results show that the proposed improved model has a MOTA improvement of 2.3%to 3.3%compared to contrast models.The detection speed is faster while the accuracy is close to the latest model,basically meeting the real-time requirements;The comprehensive evaluation index F value can reach over 95%in various scenarios,and the false detection rate in various actual scene detection is less than 0.3%.