A Practical Approach to detect Indoor and Outdoor Scene Recognition

In computer vision recognition of scenes is a long-time research problem. Scene can be defined as the real-time environment view which consists of a lot of views (like road, tree, building, parks, etc.) in a meaningful manner. The problem of scene recognition can be explained as assessment of labels such as "road", "building", "hall", "bedroom" or in an extra simplified way "Indoor scene" and "Outdoor scenes is classified on an input image based on the object or environment of the image. A huge amount of data is created and available every second in this growing era of digital data. Scene recognition is still a rising area that did not attain much success as compared to image recognition due to the vast variability of features in the scenic environment. Because of this reason, there is not a lot of work being completed these days in this area. This project focuses on the evaluation of problem statements using all the literature surveys completed lately and offering solutions for that problem statement. For resolving the proposed problem declaration ResNet and VGG variants are used ResNet18, 50 and 152 & VGG16, 19, and VGG19 with batch normalization are implemented. The entire scene recognition procedure is discussed in this report and the main motive is to form a foundation that can be further continued in proposing a new algorithm. Before deep learning the design and implementation of the scene recognition model depended on the low dimensional portrayal of the scene. The utilization of deep learning especially CNN for scene recognition has gotten extraordinary attention from the computer vision community.