Deep Learning Approach for Human Tracking by Utilizing Convolutional Neural Network and Long Short-Term Memory

Kenny Susanto (President University)
Rila Mandala (Institut Teknologi Bandung)

Journal of Innovation Information Technology and Application (JINITA)
Vol 8 No 1 (2026): JINITA, June 2026

The world of digitalization is evolving at a fast pace, and automation growth in large sectors of industry, as well as in all aspects of life, creates a necessity for a robust and dependable product in tracking objects. Any pre-recorded or live-stream video can benefit from the ability to track an object in it to increase its value. This research explores a new approach to building a human-tracking model using deep learning architectures such as CNNs and LSTMs. It aims to assess the deep learning capability in handling tasks compared to the baseline algorithm GOTURN. The proposed model is constructed using a combination of CNN+LSTM and a secondary model built on LSTM. The dataset is gathered from a publicly available resource, LaSOT (Large Single Object Tracking), which specializes in collecting objects for object tracking operations. The proposed model achieves higher accuracy than the baseline model, with up to a 100% difference. The result of this research may bring new insight to the rather unpopular field of computer vision in tracking objects. The development of this research will be discussed further in this paper.