针对高速公路服务区人车交互频繁、跨类别跟踪一致性不足，导致小客车载客系数检测困难，检测精度低、鲁棒性较差等问题，提出一种基于多维特征融合的小客车载客系数检测方法，以提高服务区运行数据的准确性。首先，采用YOLOv10x和ByteTrack作为检测跟踪模块，构建融合多维特征的检测系统。其次，采用卡尔曼滤波法预测目标运动轨迹，建立目标运动先验，引入分类别特征提取方法分别提取车辆与行人的外观语义特征；设计基于检测置信度的自适应加权关联策略，进行多维特征融合，并利用匈牙利算法实现目标重识别与匹配。最后，基于服务区监控视频进行载客系数检测，通过跟踪精度、跟踪精确度、ID跳变次数以及帧率等指标对算法进行对比分析。实验结果表明：基于多维特征融合的检测方法能够对载客系数进行精确统计，与传统ByteTrack仅采用IOU匹配实现目标跟踪相比跟踪精度和跟踪精确度分别提升了0.99%和16.09%，ID跳变次数减少了151个，降幅为18.33%；该方法在保证推理速度的同时，显著提升了复杂交通场景下的精度与鲁棒性，能够弥补服务区中载客系数统计的缺失，为提高服务区的运营管理与服务效能提供了理论基础，同时为推算高速公路上的实载率提供参考。

In view of the frequent interaction between pedestrians and vehicles, and the lack of consistency in cross category tracking in expressway service areas, which leads to difficulties in passenger coefficient detection, low detection accuracy and poor robustness, a passenger coefficient of passenger cars detection method based on multi-dimensional feature fusion is proposed to improve the accuracy of service area operation data. Firstly, YOLOv10x and ByteTrack are used as detection and tracking modules to build a detection system incorporating multi-dimensional features. Secondly, the Kalman filter method is used to predict the target trajectory and establish the target motion prior, and the classification feature extraction method is introduced to extract the appearance semantic features of vehicles and pedestrians respectively; An adaptive weighted association strategy based on detection confidence is designed, and multi-dimensional feature fusion is carried out. The Hungarian Algorithm is utilized to realize target re-identification(ReID) and matching. Finally, the passenger coefficient is detected based on the monitoring video in the service area, and the algorithm is compared and analyzed by indicators such as tracking accuracy, tracking precision, ID switches, and frame rate. The experimental results show that: the detection method based on multi-dimensional feature fusion can accurately count the passenger coefficient. Compared with the traditional ByteTrack algorithm that relies solely on IOU matching for object tracking,its tracking accuracy and tracking precision are improved by 0.99% and 16.09%, respectively, and the number of ID switches is reduced by 151, representing a 18.33% reduction. This method can significantly improve accuracy and robustness in complex traffic scenes while ensuring inference speed, and can make up for the lack of passenger coefficient statistics in the service area, which can provide a theoretical basis for improving the operation management and service efficiency of the service area, and provide a reference for calculating the actual loading rate on the expressway.