Unsupervised deep learning

I need an experienced deep learning professional to build an unsupervised deep learning model that uses a convolutional neural network to process videos as input and targets tracking as its output. I am specifically looking for an approach that is better equipped to handle large-scale changes between frames than traditional methods. The model should be able to identify complex relationships between frames, providing a more accurate and detailed understanding of the trajectory of an object within a video. I need the model to be implemented using the latest techniques and technologies in order to ensure its accuracy and effectiveness, and I will expect regular progress updates throughout the duration of the project.