A Review of RGB-D Image Classification Methods

Depth image popularity is continuously increasing in a few decades with successful implementation in various robotics applications. The use of depth RGB images improves the accuracy of classification methodologies. However, there are many challenges associated with depth RGB image classification methods. A systematic review of the use of depth RGB images in classification tasks may provide the researchers the insight into the existing methods and their limitations which may be useful for further development. Therefore, this paper explores the various existing approaches for image vector representation from depth RGB images in classification tasks. The existing depth RGB image classification approaches may be divided into hand-crafted, learning-based, and middle-level methods. As compared to hand-crafted and learning-based methods, middle-level feature-based methods are computationally inexpensive and are carried out with insufficient availability of high computing resources. Therefore, further study is accomplished in the view of approaches associated with this category. In addition to these methods, this review covers available indoor RGB-D datasets and different quantitative evaluation parameters.