Autonomous vehicles (AVs) depend on precise and efficient object detection (OD) for safe navigation in complex and dynamic environments. Traditional Convolutional Neural Networks (CNNs) excel at extracting local features but face limitations in capturing long-range dependencies, leading to challenges in scenarios involving occlusion, varying lighting, and diverse object scales. This paper proposes HCNN-TMOD, a hybrid framework that combines CNNs and Transformer Models (TM) to overcome these challenges and enhance object detection (OD) accuracy and speed for real-time autonomous vehicle applications. HCNN-TMOD utilizes CNNs for robust local feature extraction and TMs for capturing global contextual relationships. A feature fusion mechanism integrates outputs from both architectures, enabling improved spatial and semantic representations. The system is optimized for latency and hardware constraints and evaluated on various datasets like vehicle, pedestrians and traffic light detection, demonstrating suitability for real-world AV scenarios. Results show a 15% improvement in mean Average Precision (mAP) and a 20% reduction in detection latency compared to traditional CNN-based approaches. HCNN-TMOD performs exceptionally well in challenging conditions such as occlusion and low-light environments. The integration of CNNs and Transformers in this hybrid approach provides a significant advancement in OD for AVs, paving the way for safer, more reliable, and efficient real-time navigation systems.

Keywords :

IoT, Waste Management, Smart Campuses, Genetic Algorithms, Reinforcement Learning, Optimization, Real-Time Monitoring.