The Model and Training Algorithm of Compact Drone Autonomous Visual Navigation System

Abstract

Trainable visual navigation systems based on deep learning demonstrate potential for robustness of onboard camera parameters and challenging environment. However, a deep model requires substantial computational resources and large labelled training sets for successful training. Implementation of the autonomous navigation and training-based fast adaptation to the new environment for a compact drone is a complicated task. The article describes an original model and training algorithms adapted to the limited volume of labelled training set and constrained computational resource. This model consists of a convolutional neural network for visual feature extraction, extreme-learning machine for estimating the position displacement and boosted information-extreme classifier for obstacle prediction. To perform unsupervised training of the convolution filters with a growing sparse-coding neural gas algorithm, supervised learning algorithms to construct the decision rules with simulated annealing search algorithm used for finetuning are proposed. The use of complex criterion for parameter optimization of the feature extractor model is considered. The resulting approach performs better trajectory reconstruction than the well-known ORB-SLAM. In particular, for sequence 7 from the KITTI dataset, the translation error is reduced by nearly 65.6% under the frame rate 10 frame per second. Besides, testing on the independent TUM sequence shot outdoors produces a translation error not exceeding 6% and a rotation error not exceeding 3.68 degrees per 100 m. Testing was carried out on the Raspberry Pi 3+ single-board computer.

Unmanned aerial vehicle for internet of everything: Opportunities and challenges

“…Therefore, it is necessary to design portable or tiny AI models dedicated for UAVs. For example, there are few studies on investigating low-complex algorithms for UAVs' navigation [228], pathfinding [229], and moving-targets' tracking [188]. The future IoE applications require lightweight AI algorithms to support autonomous intelligence with quick and precise responses.…”

Gem5-X

“…Video analytics application composed of two kernels running concurrently, i.e., video encoding using Kvazaar [60] and image classification/detection using CNNs. [35,57] 640 × 480 5 Drone Navigation [25,36] 300 × 200 10 Autonomous Driving and ADAS [45,46] 640 × 480 15 exploit and demonstrate the capabilities of gem5-X in optimizing various memory and compute sub-systems to have an overall optimized architecture.…”

“…As previously mentioned, real-time video analytics, dubbed as the next-generation "killer-app" [2], is used in a wide range of domains, from video surveillance for security and monitoring, safety on construction sites, traffic monitoring and thermal monitoring to identify patients with fever, as well as for fire hazards [35,57]. Real-time video analytics is also used for autonomous drone navigation and rescue drone missions, e.g., during earthquakes, floods, or rescuing people at sea [25,36]. Autonomous drones are also being deployed for deliveries of parcels by many companies around the globe.…”

Research on vehicle attitude and heading reference system based on multi-sensor information fusion

“…1,2 Motion sensing technology based on laser radar and image fusion can provide high system accuracy for drones, which is expensive and susceptible to interference. 3 The inertial/magnetic sensor is a stand-alone and effective sensor that can be used in unconstrained environments and motions. It does not require external resources to interact and is low cost.…”