The project aimed to track the motion of any object as specified by the user and recognize the shapes drawn by him/her. Tracking was achieved in two different ways. OpenCV's Lucas Canade optical flow algorithm was used for tracking in unpredictable lighting conditions, whereas HSV color filtering tracking was used for more reliable lighting conditions. Lucas Canade optical flow algorithm's main drawback was that objects must be moved slowly; otherwise, the tracker would lose it. The image obtained was then saved and passed to a neural network. In deep learning, ResNet was used for classification. Transfer learning and data augmentation were used to overcome the problem of lack of training data.
Lucas Kanade based object tracking
HSV (Color) based object tracking
Deep Learning Framework: Keras