Object Detection at Sea Using Ensemble Methods Across Spectral Ranges. Issue 16 (2021)

• Record Type: Journal Article
• Authors: Schöller, Frederik E.T.; Plenge-Feidenhans'l, Martin K.; Stets, Jonathan D.; Blanke, Mogens
• Is Part Of: IFAC-PapersOnLine. Volume 54:Issue 16(2021)
• Page Start: 1
• Page End: 6
• Publication Date: 2021
• Subjects: Multi-modal sensor fusion -- Deep learning -- Detection performance -- Object detection at sea -- Computer vision -- Autonomous Marine Vehicles; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2021.10.065
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Having the option of a temporally unmanned bridge when conditions allow, while maintaining or even enhancing navigational safety, is a long term aim in the maritime industry. Such a system requires excellent perception of the environment using an array of sensors. This paper investigates performance of object detection at sea using electro-optical sensors in relevant spectral ranges and discusses how missed detection risk is minimised for objects within navigation range. Using a combination of cameras in visible, near- and far infrared ranges, convolutional neural networks are employed for object detection. Ensemble techniques are suggested to minimise the amount of missed detections and it is shown how optimisation of confidence thresholds can be used to increase performance. The results are based on image data from vessels in near-coast operation in Danish waters.