Nighttime Traffic Object Detection via Adaptively Integrating Event and Frame Domains

•The first fusion object detection framework for nighttime traffic scenarios permits the accurate detection of vehicles and pedestrians through the adaptive fusion of events and RGB sensors.•A novel learnable adaptive selection and fusion module (ASFM) can effectively solve the problem of imbalanced multimodal learning of event and frame branches in various dynamic scenes and enhance the performance of fused features.•A novel multimodal attention fusion multi-level feature pyramid network (MAF-FPN) additionally extracts depth features, which ensures the accuracy of the detection head network.•A real-world nighttime traffic scene detection dataset, dubbed NightDVS22, contains RGB frames of various traffic scenes, the corresponding event streams, and high-quality manual annotation labels.