Traction force plays a critical role in the design of off-road vehicles and the evaluation of terrain trafficability. The accuracy of the traditional Bekker-Wong semi-empirical formula for predicting wheel traction force depends on the precision of parameters such as the entry contact angle, departure contact angle, and maximum stress angle. However, the calculation of these angles is often influenced by multiple empirical factors, which significantly reduces the accuracy of traction force prediction.To improve prediction accuracy, this study proposes a hybrid traction force prediction method that integrates Abaqus simulation with a semi-empirical model. First, a wheel-soil interaction simulation model was established in Abaqus using the FEM-SPH coupling method. Then, computer vision-based image feature extraction techniques were applied to process the simulation result contours, obtaining accurate values for the entry contact angle, departure contact angle, and maximum stress angle. These values were substituted into the Bekker-Wong semi-empirical formula, effectively overcoming the heavy reliance on empirical parameters inherent in traditional methods.To verify the accuracy and applicability in full-vehicle scenarios of the hybrid method, the traction force under medium and low slip conditions in wet, soft clay was predicted using the proposed approach. Meanwhile, the traditional semi-empirical formula method was employed to calculate the contact angles based on settlement and empirical parameters, and thereby compute the traction force. Traction forces under various working conditions were also obtained through high-fidelity simulation of off-road vehicle driving on soft soil surfaces using the Chrono platform. By comparing the results from the empirical formula method, the hybrid method, and the high-fidelity simulation, the effectiveness and accuracy of the hybrid method were validated under complex environments involving multi-wheel coupling and soil disturbance.