This paper presents an extension to the Simple Online and Realtime Tracking (SORT) algorithm, which improves performance by incorporating appearance information through a pre-trained deep association metric. The original SORT algorithm uses a simple association metric based on bounding box overlap, which is effective for short-term tracking but struggles with long occlusions. The proposed method integrates a deep neural network trained on a large-scale person re-identification dataset to enhance the association metric, combining motion and appearance information. This allows the tracker to maintain object identities through longer periods of occlusion, reducing the number of identity switches by 45%. The method uses a weighted combination of two metrics: the Mahalanobis distance for motion information and a cosine distance for appearance information. The association problem is solved using a matching cascade that prioritizes tracks with smaller age, i.e., more recent tracks. The method is evaluated on the MOT16 benchmark, showing improved performance in terms of tracking accuracy and robustness against occlusions and misses. The system runs efficiently on a modern GPU, achieving real-time performance with a frame rate of approximately 20 Hz. The proposed method is a strong competitor to other online tracking algorithms, offering competitive performance while maintaining simplicity and real-time operation.This paper presents an extension to the Simple Online and Realtime Tracking (SORT) algorithm, which improves performance by incorporating appearance information through a pre-trained deep association metric. The original SORT algorithm uses a simple association metric based on bounding box overlap, which is effective for short-term tracking but struggles with long occlusions. The proposed method integrates a deep neural network trained on a large-scale person re-identification dataset to enhance the association metric, combining motion and appearance information. This allows the tracker to maintain object identities through longer periods of occlusion, reducing the number of identity switches by 45%. The method uses a weighted combination of two metrics: the Mahalanobis distance for motion information and a cosine distance for appearance information. The association problem is solved using a matching cascade that prioritizes tracks with smaller age, i.e., more recent tracks. The method is evaluated on the MOT16 benchmark, showing improved performance in terms of tracking accuracy and robustness against occlusions and misses. The system runs efficiently on a modern GPU, achieving real-time performance with a frame rate of approximately 20 Hz. The proposed method is a strong competitor to other online tracking algorithms, offering competitive performance while maintaining simplicity and real-time operation.