This paper proposes an orthogonal omnidirectional wheeled two-dimensional conveyor platform and the inverse kinematic equations based on the cell model and the omnidirectional wheel model are derived, respectively. The two models use the way of the circumcircle of the cargo box and the dot product method to obtain the cell or omnidirectional wheel directly contacted under the cargo box, respectively. This paper uses YOLOV8_OBB and Strong SORT target tracking algorithms to assign tracking IDs to the cargo box and output the box position information. A feedback mechanism for faulty omnidirectional wheels or faulty cells is proposed, where the faulty area is classified as impassable during path planning and the A* path planning algorithm is used for path planning to avoid the faulty area. A visual feedback-based path tracking is adopted to enable timely correction when the cargo box deviates from the established path. The experimental results show that the omnidirectional wheel-based model improves the path-tracking accuracy by 44.52% and reduces the average power consumption by 2.6 times compared to the cell-based model.