Recent advancements in monitoring systems for the elderly include the widespread use of fisheye lenses and omnidirectional cameras. Multi-object tracking (MOT) methods have been developed for automated monitoring, particularly those using Convolutional Neural Networks (CNNs) like DeepSORT. DeepSORT, known for its accuracy and speed, uses CNNs to identify features of tracked objects. However, applying DeepSORT directly to footage from omnidirectional cameras can lead to reduced tracking accuracy. This paper proposed a new MOT method based on DeepSORT, designed explicitly for omnidirectional images, suitable for broad indoor areas. It improves performance by adjusting Kalman filter parameters and the threshold for the cost matrix application. This new method outperforms DeepSORT in all metrics.