Single-Image Insect Pose Estimation by Graph Based Geometric Models and Random Forests
We propose a new method for detailed insect pose estimation, which aims to detect landmarks as the tips of an insect’s antennae and mouthparts from a single image. In this paper, we formulate this problem as inferring a mapping from the appearance of an i
- PDF / 3,120,048 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 98 Downloads / 169 Views
Abstract. We propose a new method for detailed insect pose estimation, which aims to detect landmarks as the tips of an insect’s antennae and mouthparts from a single image. In this paper, we formulate this problem as inferring a mapping from the appearance of an insect to its corresponding pose. We present a unified framework that jointly learns a mapping from the local appearance (image patch) and the global anatomical structure (silhouette) of an insect to its corresponding pose. Our main contribution is that we propose a data driven approach to learn the geometric prior for modeling various insect appearance. Combined with the discriminative power of Random Forests (RF) model, our method achieves high precision of landmark localization. This approach is evaluated using three challenging datasets of insects which we make publicly available. Experiments show that it achieves improvement over the traditional RF regression method, and comparably precision to human annotators.
Keywords: Insect pose estimation forest
1
·
Landmark detection
·
Random
Introduction
Automated image based tracking and pose estimation receives increasing interests of both biology and computer science community, as its developments enable remotely quantify and understand individual behavior previously impossible [1]. Therefore, automatic insect tracking techniques have been an research topic in biological image analysis [2–4]. The movements of harnessed insects’ bodyparts, such as antennae or mouthparts, provide information for behavioral study. Motivated by latest behavioral studies in biology [5], we aim to localize the landmark as the tips of bodyparts (e.g. a bee’s antennae or tongue shown in Fig. 1) to provide detailed pose information. In contrast to most existing works that aim at estimating the center of mass (position), detecting the detailed body posture and position of appendages (pose) is more challenging. Most existing tracking/pose estimation algorithms are not applicable for our task, due to a number of specific challenges and constraints: c Springer International Publishing Switzerland 2016 G. Hua and H. J´ egou (Eds.): ECCV 2016 Workshops, Part I, LNCS 9913, pp. 217–230, 2016. DOI: 10.1007/978-3-319-46604-0 16
218
M. Shen et al.
– Occlusion. Insect body parts are highly clustered due to their small sizes, thus self occlusions are prevalent in insect body parts (Fig. 1b). As a result, it is difficult to estimate the pose. – Unstructured appearance. Insect body parts have dark appearance, similar shape and no texture. Moreover, our image data is a set of 2D videos, and does not contain depth information. Their unstructured appearance makes them difficult to differentiate. – Complex motion. A varying number of body parts are observed in consecutive video frames (e.g. the bee tongue does not appear in Fig. 1c), thus we have incoherent motion paths and the trajectories have long tracking gaps. Motion cues furthermore provide only little information to predict the current pose. In the videos to be analyzed, the insect will
Data Loading...
