Low power microcontrollers have become widely available. Hence, they have been used in several stand-alone applications in which the developed system depends on battery or energy harvesting module. One such application is surveillance aiming to observe a selected region or target in time. Due to the complexity of the problem and real-time constraints in operation, several object trackers have been proposed in literature. An object tracker produces the trajectory of an object from a given image sequence. To do so, two major steps are taken as object representation and trajectory prediction. Here, the computation load for tracking and object representation strength plays adversary effects most of the time. Moreover, the overall system to be deployed in a remote location casts serious limitations on the tracking method to be used. Therefore, we propose a probabilistic object representation-based object tracking method to work on low power Arm Cortex-M4 and -M7 core microcontrollers in this study. The proposed method aims to represent the object to be tracked as simple as possible. On the other hand, the method provides an effective way of describing the object to be tracked. Therefore, the novelty of the proposed method is adding a simple yet flexible probabilistic object representation method to the tracking framework. The probabilistic object representation method can be easily merged with the Bayesian framework which is extensively used in trajectory prediction. To do so, we use the particle filter based Bayesian tracking method. As we form the overall system for object tracking, we compare it with similar methods in the literature under real-time constraints. We provide experimental results to show the strengths and weaknesses of the proposed method in comparison with the existing ones.

低功耗微控制器已变得广泛可用。因此，它们已用于多个独立应用，其中开发的系统依赖于电池或能量收集模块。一种此类应用是旨在及时观察选定区域或目标的监视。由于问题的复杂性和操作中的实时限制，文献中已经提出了几种对象跟踪器。对象跟踪器从给定的图像序列中生成对象的轨迹。为此，采取了两个主要步骤，即对象表示和轨迹预测。在这里，跟踪和对象表示强度的计算负载在大多数情况下会起到对抗作用。此外，要部署在远程位置的整个系统严重限制了要使用的跟踪方法。所以，在本研究中，我们提出了一种基于概率对象表示的对象跟踪方法，以在低功耗 Arm Cortex-M4 和 -M7 内核微控制器上工作。所提出的方法旨在尽可能简单地表示要跟踪的对象。另一方面，该方法提供了一种描述待跟踪对象的有效方式。因此，所提出方法的新颖之处在于向跟踪框架添加了一种简单而灵活的概率对象表示方法。概率对象表示方法可以很容易地与广泛用于轨迹预测的贝叶斯框架合并。为此，我们使用基于粒子滤波器的贝叶斯跟踪方法。当我们形成用于对象跟踪的整体系统时，我们将其与实时约束下的文献中的类似方法进行比较。