Real-Time Object Detection and Classification of Small and Similar Figures in Image Processing

Abstract

In the current work we present an image processing architecture for real time object detection and classification. We use a combination of the widely known techniques YOLO v2 and Convolutional Neural Network classifiers, obtaining great improvements in the detection level with a minimum loss of performance compared to YOLO v2. We apply this technique in a domain where the objects to be detected are like each other and occupy small areas in the images, as it occurs with video traffic signs domain. With this approach, we achieve real-time video processing capabilities for a test set of 10 different signs classes. The tests results achieved process time levels faster than widely recognized algorithms, such as Fast R- CNN and Faster R-CNN, so it allows to project its use in real- time object detection.