Real-Time Object Detection & Tracking System (Faster R-CNN & RNN)

Faster is an advanced computer vision system that uses a hybrid approach combining R-CNN and RNN architectures to perform highly accurate object detection and tracking in video streams.

This project develops a hybrid deep learning model for real-time object detection and tracking in dynamic environments. The study aims not only to find objects (detection) but also to analyze temporal dependencies within the video stream. Technical Methodology and Architecture: Faster R-CNN (Region-based Convolutional Neural Networks): This Region Recommendation Network (RPN) based architecture was used to determine and classify the location of objects in the image. High mAP (mean Average Precision) scores were targeted. RNN (Recurrent Neural Networks): Integrated into the system to predict the trajectory of objects by establishing the temporal relationship between video frames and to maintain tracking during momentary occlusions. Dataset and Training: The model was trained on standard datasets such as COCO and Pascal VOC and optimized using transfer learning methods. Performance: Real-time FPS values ​​were achieved on the GPU using CUDA acceleration.