Multi Fish Detection and Tracking Using RGB-D Vector

摘要

Multi fish detection and tracking has got cogent significance in the area of image processing. The tracking methodologies has improved tremendously during the last decades, however, it is still a complex problem to solve. The importance of object tracking can be seen in various applications like video communication, machine inspection, robot navigation and augmented reality. This paper presents Microsoft Kinect camera sensor-based fish detection and tracking. The Kinect sensor provides depth and RGB images along with reduced noises and holes. The admiring traits of the depth and color information given by the Kinect sensor brings up various opportunities to resolve elementary issues in the field of computer vision. Fish detection and tracking in a constrained environment (aquarium) aids in the research of marine biologists to understand and study the behaviour of fishes. The use of commodity hardware devices like Microsoft Kinect for detection and tracking is economical for the researchers and also helps to automate the study of fishes round the clock without any requirement of manual intervention. In this paper, we have calibrated depth and RGB images to detect and track multiple fishes present in each frame of a recorded video. Due to the bleak body deformation of fish that results in their complex motion and multiple occlusions, robust fish tracking from video image sequences is a highly challenging task. In this paper a methodology is proposed to overcome these problems and multiple fishes can be tracked using centroid of the fish as central component and also the detection accuracy for RGB and RGBD frames are compared. The results show that the performance of RGBD method has 12.51% more accuracy on average over RGB based input.