Autonomous driving faces the long-tail problem of vehicle perception, risking perception failure in certain scenarios. Vehicle-infrastructure collaborative perception shows potential in addressing this issue, yet technical bottlenecks persist. To address the asynchrony in the temporal dimension and the inconsistency in the spatial dimension of vehicle-infrastructure perception information. A late fusion method with spatial-temporal alignment is proposed. After converting the roadside results to coordinates, time compensation is performed based on a uniform motion model. An improved Iterative Closest Point (ICP) method is used to calculate the transformation matrix and finely adjust the roadside results. Finally, the perception results from both the vehicle and infrastructure are fused. Compared to vehicle perception and roadside perception, the vehicle-infrastructure collaborative perception method improved the accuracy of the bird’s-eye view by 44% and 9.4%, respectively, validating the positive role of vehicle-infrastructure collaborative perception in enhancing vehicle perception capabilities. Furthermore, compared with traditional late-fusion methods, the proposed method in this paper achieves significantly higher bird’s-eye-view precision under various vehicle-infrastructure asynchrony delays, effectively addressing the issue of spatiotemporal inconsistency in vehicle-infrastructure perception results.