AHOD: adaptive hybrid object detector for context-aware and real-time object detection in complex environments

This article introduces AHOD (Adaptive Hybrid Object Detector), a novel architecture designed to balance inference speed, detection accuracy, and contextual adaptability. AHOD integrates three innovations: (1) Feature Pyramid Enhancement (FPE) with spatio-temporal attention for robust multi-scale detection; (2) Dynamic Context Module (DCM) that adjusts thresholds based on scene semantics; and (3) Fast and Accurate Detection Head (FADH) that distributes computational load across CPU and GPU to minimize latency. Experiments on COCO, Pascal VOC, and BDD100K demonstrate that AHOD achieves an mAP of 91.4%, surpassing YOLOv8 and Faster R-CNN by over 7% while reducing inference time by 30%. Ablation studies validate each module's contribution, positioning AHOD as a promising solution for real-time, context-sensitive applications.