Deep Learning with Image Processing for Object Detection in Traffic Cameras

Traffic surveillance is a key factor in ITS whereby accurate and real-time object detection assures improvement of road safety and traffic management. This paper advances a deep-learning-based perspective that combines image processing techniques with convolutional neural networks (CNNs) to maximize the object detection accuracy over the traffic camera feed. The filtering, edge detection, and feature extraction of images for data preprocessing enhance the model performance. This setup provides high inference speed with reliable detection of vehicles via YOLO (You Only Look Once) and Faster R-CNN. The obtained experimental results from our evaluation show great enhancement in the detection accuracy which places the model in a position for real-time application in traffic surveillance. The study also investigates the computational efficiency and real-life boundaries for implementing this strategy, providing an exhaustive account of the suitability of the proposed method in larger-scale applications in ITS.