Part-based multi-task deep network for autonomous indoor drone navigation

Abstract

Two common methods exist for solving indoor autonomous navigation and obstacle-avoidance problems using monocular vision: the traditional simultaneous localization and mapping (SLAM) method, which requires complex hardware, heavy calculations, and is prone to errors in low texture or dynamic environments; and deep-learning algorithms, which use the fully connected layer for classification or regression, resulting in more model parameters and easy over-fitting. Among the latter ones, the most advanced indoor navigation algorithm divides a single image frame into multiple parts for prediction, resulting in doubled reasoning time. To solve these problems, we propose a multi-task deep network based on feature map region division for monocular indoor autonomous navigation. We divide the feature map instead of the original image to avoid repeated information processing. To reduce model parameters, we use convolution instead of the fully connected layer to predict the navigable probability of the left, middle, and right parts. We propose that the linear velocity is determined by combining three prediction probabilities to reduce collision risk. Experimental evaluation shows that the proposed method is nine times smaller than the previous state-of-the-art methods; further, its processing speed and navigation capability increase more than five and 1.6 times, respectively.

FADM-SLAM: a fast and accurate dynamic intelligent motion SLAM for autonomous robot exploration involving movable objects

“…The technique analyzes the object’s motion in the image and then determines the correlation between the feature points in the current and the previous frames. Using the FAST algorithm, this method gathers trustworthy key points (such as FAST; Zhang et al , 2020) when necessary. The appropriate matching points are obtained by minimizing the photometric error under the assumption of a constant gray value, and the pose is solved using matching point pairs.…”

“…Bescos et al (2018) describe DynaSLAM, a visual SLAM system, that extends ORB-SLAM2 (Mur-Artal and Tardós, 2017) with dynamic object identification and backdrop inpainting. A novel semantic SLAM system for RGB-D cameras is described that detects possibly moving objects using Mask R-CNN to achieve resilience in dynamic settings (Zhang et al , 2020). Cui and Ma (2020) introduce SDF-SLAM: semantic depth filter SLAM, a visual semantic SLAM system for dynamic environments that uses depth filter technology to determine if a 3D map point is dynamic or not.…”

Preliminary Study on Drone Navigation in Urban Environments using Visual Odometry and Partially Observable Monte Carlo Planning

“…A type of deep learning approach is applied to the large amount of crash video footage captured by the front cameras of drones. Literatures [15] and [16] both use deep learning techniques to learn from the video dataset and execute appropriate actions based on the video feed. Based on the experiment results from both articles, demonstrates that a drone can navigate in an unknown environment with an unknown position for the destination using computationalintelligence-assisted visual navigation.…”

Scene terrain classification for autonomous vehicle navigation based on semantic segmentation method

“…To overcome and increase the model accuracy in low-cost object identification, an AV system with monocular vision sensor-based scene and terrain segmentation using the Mask R-CNN method is proposed in this paper. A single image is converted into thousands of images augmented dataset (Zhang et al, 2020), as like a thousand of image samples are augmented and analysed using U-Net and Mask R-CNN, and the results are compared to see which performs better in detecting topography and obstacles in low-quality image samples. With low temporal latency and high model accuracy, the suggested work focuses on a vision sensor data collection system to segregate the terrain and the obstacle.…”