Undergraduate research.
We present a multi-object tracking system to track small insects such as ants and bees. Motion-based object tracking recognizes the movements of objects in videos using information extracted from the given video frames. We applied several computer vision techniques, such as blob detection and appearance matching, to track ants. Moreover, we discussed different object detection methodologies and investigated the various challenges of object detection, such as illumination variations and blob merge/split. The proposed system effectively tracked multiple objects in various environments.
Automatic motion tracking system for analysis of insect behavior
1. Automatic Motion Tracking System for
Analysis of Insect Behavior
Darrin Gladman, Jehu Osegbe, Wookjin Choi, and Joon Suk Lee
Virginia State University, 1 Hayden St, Petersburg, VA, USA
METHODS
INTRODUCTION
RESULTS
1. “ANTS–ant detection and tracking,” (oct 2019).
2. Bernardin, K. and Stiefelhagen, R., “Evaluating Multiple Object
Tracking Performance: The CLEAR MOT Metrics,” EURASIP Journal on
Image and Video Processing (2008).
CONCLUSIONS
REFERENCES
Computer vision is a field of computer science that allows
computers to see, identify, and process images in the same
way as human vision, and then provides useful information
based on the observation. Insects behavior studies enable us
to understand the division of the social insect group and the
specialization of tasks that can benefit modern
applications, such as wireless communication.
This paper aims to explore the different methodologies of
object detection and tracking while also reviewing the various
challenges and aspects of detecting ants and other insects. .
Tracking an individual insect in a group for a long period is the
usual method to study their behavior. However, this method is
prone to human error and is labor-intensive. This paper aims
to resolve these problems and offers a solution to track
unlabeled ant individuals automatically. Therefore, we
introduce an online MOT framework to track ants, which
provides several evaluation measures such as precision and
running time.
To track the ants in the video a multi- object tracker was
implemented. A pre-trained ResNet model was employed to
extract the features from the detected ROIs for more precise
object detection. The tracker matched ant locations over video
frames using cosine similarity of the features. Furthermore, a
simple blob detection method was implemented to compare
the two algorithms. To quantify the performances of the
algorithms MOT indicators were presented to further evaluate
object detection, localization, and tracking performance [2].
Tracking individuals in an insect group is a common
approach to understanding their behavior
A bar chart was created to visualize better the comparison
between the KCF algorithm and the blob detector. This is
illustrated in Figure 4, where algorithms MOTA indicators were
compared for each video. From a glance, it is seen that the
KCF algorithm and the blob detector performed about the same
in the indoor videos. This suggests that detection works well in
simple environments without much background noise about
equally. The blob detection method can only detect the regions
in a digital image that differed in properties, such as brightness
or color in the video. Since there wasn’t enough contrast
between the ant’s appearance and ground of the outdoor video,
the blob detector didn’t run accurately.
We introduced a motion tracking system to track small insects,
such as ants and bees effectively. We used a public dataset
that contained both indoor and outdoor videos. We adopt the
online MOT framework and use the standard Kalman filter to
predict the motion state of objects, to achieve real-time
tracking. We compared the blob detection and the KCF
algorithm in both indoor and outdoor videos.
Fig. 1 Process chart of the proposed method
Our result points out that the blob detector
performed worse in detecting the ants in both videos.
Fig. 2 Binary image of the Ants in Blob Detector
Our work proposes a straightforward approach for effectively
tracking the ants (Figure 1). The proposed method was
evaluated using the ANTS dataset [1], which contained a
sequence of images of ants in indoor (5 videos) and outdoor
(1 video) environments.
Fig. 3 Outdoor and Indoor frames of the blob detection
Fig. 4 MOTA Comparison of the KCF and the blob
detection algorithm