Localizing and tracking dense crowd of microbes by joint association and detection refinement

This paper presents a method for detecting and tracking large number of arbitrary-oriented and densely aggregated microbes from image sequences captured under microscope. We first propose an integral channel feature (ICF)-based detector which is able to localize the dense and arbitrarily oriented targets with low false positive rate. Then instead of treating target detection and tracking as two separated problems as many previous works did, we propose to refine the detection results in the data association process. The kinematic pattern of microbes is well modeled with the proposed integral sliding energy (ISE), which is combined with detection response in a hybrid cost function. Minimizing the cost function allows us to simultaneously select the true targets from the detections and to match the targets across two consecutive frames. Systematical experiments have been conducted to demonstrate the effectiveness of proposed method.

Automated summary

This paper presents a method for detecting and tracking a large number of densely aggregated microbes with arbitrary orientations in image sequences captured under a microscope. By proposing an ICF-based detector and refining detection results through a data association process, the kinematic pattern of microbes is accurately modeled and used to select true targets and match them across frames effectively. Experimental results demonstrate the effectiveness of the proposed method.