Mine like object detection and recognition based on intrackability and improved BOW

• Published in: 2016 IEEE International Conference on Cyber Technology in Automation, Control, and Intelligent Systems (CYBER) pp. 222 - 227
• Main Authors: Siquan Yu, Jinxin Shao, Zhi Han, Lei Gao, Yang Lin, Yandong Tang, Chengdong Wu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2016
• Subjects: Feature extraction; Image segmentation; Object detection; Sonar detection; Support vector machines; Videos
• DOI: 10.1109/CYBER.2016.7574826

In this paper, we present an automatic system of mine like object detection and recognition for sonar videos. This system is implemented with two main methods. One is the object detection and segmentation with intrackability, another is object recognition of mine like based on improved BOW algorithm and Support Vector Machine (SVM). Intrackability is defined by the concept of entropy, and can reflect the difficulty and uncertainty in tracking certain elements on the time axis. Therefore, our segmentation and detection method can effectively eliminate complex noise in sonar image to guarantee the more accurate object segmentation and detection. In our recognition method of mine like object, we use an improved BOW and SVM to implement the more accurate recognition for mine like objects. In the method, an improved BOW algorithm is utilized for image feature extraction, due to that it can represent local and global feature of image in a more comprehensive way; and then the object recognition is implemented with SVM. Our extensive experiments show that our system can accurately detect and recognize mine like objects in real-time.