Federated learning (FL) is a collaborative learning approach in which multiple clients or organizations jointly train a machine learning model without revealing their raw data.  Vertical federated learning ( VFL) is an FL approach where data is partitioned based on the feature space, with each participating client owning distinct features. Clients may or may not share the same data samples, but the features they hold are unique to their datasets.  The rapid advancements and practical applications of VFL has inspired us to leverage it for real-time object detection tasks in vehicular applications and to evaluate key performance metrics related to the learning process.This thesis focuses on analyzing various VFL techniques, performing data analysis, and evaluating object detection models in the context of vehicular applications of object detection. This thesis utilizes a 2D image dataset within the context of the vehicular network. In a VFL setting, where each client possesses distinct features, a method called split model VFL (SMVFL) has been proposed to support this functionality using the 2D image dataset. In this approach, the object detection model is divided into a feature extraction layer and a detection layer. Within the VFL framework, only one active client will have access to the detection layer to perform object detection, while multiple clients will utilize the feature extraction layer. The parameters and gradients from these layers are aggregated on a central server to evaluate a global model based on the aggregated parameters. Each client independently extracts significant features using their models, and the proposed SMVFL method has been implemented and experimentally evaluated, with the results presented. The results obtained from SMVFL have been compared with other methods, including single client centralized training ( SCCT) and aggregated dataset central training (ADCT). The key findings indicate that SMVFL outperforms SCCT while providing a privacy-preserving and distributed training solution. With more communication rounds, SMVFL has the potential to achieve performance closer to ADCT.