|
Simultaneous detection and tracking of multiple moving objects in forward looking thermal IR image sequences
|
A. Abootalebi *   , F. Samadzadegan , Gh. Abdi |
|
|
|
Abstract: (7040 Views) |
| Object detection and tracking in image sequences is a key topic in photogrammetry and computer vision and can be viewed as a lower level vision tasks to achieve higher level event understanding. Images from different kind of sensors generally have different pixel characteristics due to the phenomenological differences between the image formation process of the sensors. In recent years, some approaches have already been proposed to detect and recognize moving objects in thermal Infra-Red (IR) image sequences, especially in the field of monitoring, security and surveillance. Nevertheless, tracking of multiple moving objects in forward looking thermal IR image sequences can be complex due to loss of information caused by projection of the 3D world on a 2D image, noise in images, complex object motion, non-rigid or articulated nature of objects, partial and full object occlusions, complex object shapes, scene illumination changes, and simultaneous changing of the situation of objects and sensor. This paper presents a novel method that overcomes most of the shortcomings of the existing detection and tracking algorithms in forward looking thermal IR image sequences. In this context, the Speeded-Up Robust Features (SURF) detector, Nearest Neighbor (NN), and RANdom SAmple Consensus (RANSAC) have been used to eliminate motion induced by the motion of the platform. Next, the Accumulative Frame Difference (AFD) has been used to detect moving objects from the ego-motion compensated input sequences. Also, an outlier removal algorithm based on Mean Gray Area (MGA), compactness, and eccentricity has been applied to detect and remove non-moving objects induced by errors in alignment, parallax, and etc. Finally, a tracking algorithm based on a constant velocity motion model, and various cues for object correspondence has been applied to perform tracking moving objects using their motion histories. The potential of the proposed method was evaluated through comprehensive experimental tests conducted on a wide variety of datasets. We compare the performance of our detection and tracking setup against different evaluation metrics, namely Hit Rate (HR), False Alarm Rate (FAR), Multi Object Detection Precision (MODP), and Multi Object Tracking Precision (MOTP) for a subset of ten sequences from our datasets. Inspecting the results, the proposed method has the potential to track moving of different pedestrians and cars objects with different motion characteristics effectively and efficiently. |
|
| Keywords: Thermal IR, Object Detection, Object Tracking, Image Sequences |
|
|
Full-Text [PDF 1967 kb]
(1813 Downloads)
|
Type of Study: Research |
Subject:
Photo&RS
Received: 2015/06/13
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
