With recent active research related to autonomous driving, object tracking technology using autonomous driving sensors such as LiDAR and radar has also undergone extensive development. Accordingly, attempts are being made to apply autonomous sensors not only on autonomous vehicles but also in various fields such as security and surveillance. However, since security and surveillance systems should be able to detect and track objects even under extreme environmental conditions such as snow, rain, and fog during the day or night, radar systems that meet the relevant requirements are essential. In South Korea, the distance of the Military Demarcation Line (MDL) is 250 km, and a considerable investment would be required to install more than 1,000 radars and PCs with built-in GPUs in all sections for a border security and surveillance system. Therefore, in this study, a Kalman filter-based object tracking system is explored rather than applying deep learning, which requires GPU processing. Additionally, most objects along the MDL are highly likely to be suspicious objects, so a radar sensor is most suitable because it provides coordinates, distance, and speed of movement without needlessly determining whether an object is an enemy or not. For accurate object detection and tracking performance, two motion models for the Kalman filter, a constant acceleration model (CAM) and a constant turn rate and acceleration model (CTRAM), are compared to identify a suitable model for each object movement state.