HHA-based CNN image features for indoor loop closure detection

Loop closure detection is an important part of visual simultaneous location and mapping (SLAM) system. Most of traditional loop closure detection approaches using hand-crafted features often lack robustness with respect to object occlusions and illumination changes, especially for the complicated indoor environment. Recently, convolutional neural network (CNN) makes a huge impact on many computer vision and pattern recognition applications. Depth images have complementary information to RGB images, which can encode the structural information of the scene. With the availability of inexpensive RGB-D sensors, it is expected that depth information can increase the accuracy in many computer vision applications. In this paper, we focus on indoor loop closure detection using the depth information. For the first time, we introduce a simple CNN model to train the depth images encoded by HHA method for indoor loop closure detection. The experiment demonstrates that HHA based CNN features can fully utilize the structural information of the scene, which shows that it is preferable than using raw depth images for loop closure detection.