This paper presents a novel approach to camera-based 3D object detection, a critical task in autonomous driving systems. We propose a method that leverages historical information and introduces a plug-and-play characteristic during training. Our approach builds upon existing methods by extending the concept of historical feature augmentation and fully exploiting historical information, significantly enhancing detection performance. We also introduce an auxiliary loss operating at the Bird’s Eye View (BEV) level, which meticulously cultivates the model’s environmental cognition. To address the temporal dimension’s inherent lack in the BEV framework, we incorporate a temporal embedding module into the BEV feature amalgamation. Our method elevates the detection paradigm by harnessing historical and temporal information and integrating a BEV-level supervisory framework. We validate our approach through extensive experiments, demonstrating its superiority in real-world scenarios and potential for applications in autonomous driving, video surveillance, and intelligent transportation systems.