Recently, the Internet of Vehicles (IoV) technology has played a pivotal role in enhancing transportation efficiency and safety. In this context, high-density vehicles generate more sensitive heterogeneous data, increasing privacy concerns in secure IoV data sharing. Federated Learning (FL) and Split Learning (SL) are trending paradigms of collaborative learning that facilitate potential solutions to privacy and heterogeneous data concerns. Thus, developing a privacy-preserving collaborative learning strategy is crucial for improving performance. This paper introduces a novel Privacy-Preserving Federated and Adaptive Split Learning (PPFedSL) strategy, enabling secure and efficient data sharing for IoV in smart city environments. This model integrates adaptive SL and FL by establishing a dual-tier privacy-preserved data-sharing strategy. Exploiting lightweight and hybrid cryptographic algorithms across different tiers ensures security and efficiency in data sharing across edge and cloud infrastructures without imposing significant computational overhead. This approach has designed two phases: privacy-preserving SL-enabled vehicle-edge collaboration and privacy-preserving FL-enabled edge-cloud collaboration. The proposed strategy effectively addresses latency constraints by delegating emergency decision-making at the edge level, where data is processed close to the IoV devices. Edges can inspect and respond to pressing data streams in real-time and guarantee timely interventions for latency-sensitive traffic management and collision avoidance scenarios. Finally, the experimental results demonstrate the efficiency of this proposed PPFedSL. The PPFedSL enhances the robustness efficiency by 93.2% and learning accuracy by 98.4% with high privacy preservation and heterogeneous data handling.