NITCAD - Developing an object detection, classification and stereo vision dataset for autonomous navigation in Indian roads

The advent of artificial intelligence and deep learning has provided sophisticated functionality for sensor fusion and object detection and classification which have accelerated the development of highly automated and autonomous ships as well as decision support systems for maritime navigation. It is, however, challenging to assess how the implementation of these systems affects the safety of ship operation. We propose to utilize marine training simulators to conduct controlled, repeated experiments allowing us to compare and assess how functionality for autonomous navigation and decision support affects navigation performance and safety. However, although marine training simulators are realistic to human navigators, it cannot be assumed that the simulators are sufficiently realistic for testing the object detection and classification functionality, and hence this functionality cannot be directly implemented in the simulators. We propose to overcome this challenge by utilizing Cycle-Consistent Adversarial Networks (Cycle-GANs) to transform the simulator data before object detection and classification is performed. Once object detection and classification are completed, the result is transferred back to the simulator environment. Based on this result, decision support functionality with realistic accuracy and robustness can be presented and autonomous ships can make decisions and navigate in the simulator environment.

Download Full-text

Real-time object detection and pose estimation using stereo vision. An application for a Quadrotor MAV

2015 Workshop on Research, Education and Development of Unmanned Aerial Systems (RED-UAS) ◽

10.1109/red-uas.2015.7440992 ◽

2015 ◽

Cited By ~ 3

Author(s):

Shuting Zhou ◽

Gerardo Flores ◽

Eric Bazan ◽

R. Lozano ◽

A. Rodriguez

Keyword(s):

Object Detection ◽

Real Time ◽

Pose Estimation ◽

Stereo Vision

Download Full-text

A survey on stereo vision-based autonomous navigation for multi-rotor MUAVs

Robotica ◽

10.1017/s0263574718000358 ◽

2018 ◽

Vol 36 (8) ◽

pp. 1225-1243 ◽

Cited By ~ 5

Author(s):

Jose-Pablo Sanchez-Rodriguez ◽

Alejandro Aceves-Lopez

Keyword(s):

Unmanned Aerial Vehicles ◽

Stereo Vision ◽

Autonomous Navigation ◽

Three Dimensional ◽

Field Of View ◽

Line Of Sight ◽

Onboard Computer ◽

Low Weight ◽

Vision Based Navigation ◽

Difficult Challenge

SUMMARYThis paper presents an overview of the most recent vision-based multi-rotor micro unmanned aerial vehicles (MUAVs) intended for autonomous navigation using a stereoscopic camera. Drone operation is difficult because pilots need the expertise to fly the drones. Pilots have a limited field of view, and unfortunate situations, such as loss of line of sight or collision with objects such as wires and branches, can happen. Autonomous navigation is an even more difficult challenge than remote control navigation because the drones must make decisions on their own in real time and simultaneously build maps of their surroundings if none is available. Moreover, MUAVs are limited in terms of useful payload capability and energy consumption. Therefore, a drone must be equipped with small sensors, and it must carry low weight. In addition, a drone requires a sufficiently powerful onboard computer so that it can understand its surroundings and navigate accordingly to achieve its goal safely. A stereoscopic camera is considered a suitable sensor because of its three-dimensional (3D) capabilities. Hence, a drone can perform vision-based navigation through object recognition and self-localise inside a map if one is available; otherwise, its autonomous navigation creates a simultaneous localisation and mapping problem.

Download Full-text

A 3D Object Detection Based on Multi-Modality Sensors of USV

Applied Sciences ◽

10.3390/app9030535 ◽

2019 ◽

Vol 9 (3) ◽

pp. 535 ◽

Cited By ~ 1

Author(s):

Yingying Wu ◽

Huacheng Qin ◽

Tao Liu ◽

Hao Liu ◽

Zhiqiang Wei

Keyword(s):

Object Detection ◽

Marine Environment ◽

Autonomous Navigation ◽

3D Object ◽

Detection Depth ◽

Generation Network ◽

Institute Of Technology ◽

3D Object Detection ◽

Complicated Surface ◽

Fusion Detection

Unmanned Surface Vehicles (USVs) are commonly equipped with multi-modality sensors. Fully utilized sensors could improve object detection of USVs. This could further contribute to better autonomous navigation. The purpose of this paper is to solve the problems of 3D object detection of USVs in complicated marine environment. We propose a 3D object detection Depth Neural Network based on multi-modality data of USVs. This model includes a modified Proposal Generation Network and Deep Fusion Detection Network. The Proposal Generation Network improves feature extraction. Meanwhile, the Deep Fusion Detection Network enhances the fusion performance and can achieve more accurate results of object detection. The model was tested on both the KITTI 3D object detection dataset (A project of Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago) and a self-collected offshore dataset. The model shows excellent performance in a small memory condition. The results further prove that the method based on deep learning can give good accuracy in conditions of complicated surface in marine environment.

Download Full-text