In recent times, the significance of next location prediction in different fields of application has gained the attention of investigators. Traffic flow forecasting, tracking devices development are some of the applications related to this fields. With huge directional trajectory information, researchers can predict where humans will travel up next. Humans use distinct routes to prevent obstructions and give space to other pedestrians. Researchers have an interest in predicting human future trajectories according to the previous locations. This paper suggests a combined deep learning model that may train human movement patterns and predict future trajectories. Here, three steps are involved including Preprocessing step, Feature extraction step and Future location Prediction step. In the preprocessing step, input video from the dataset is converted into the frames (images) and then filtered by an improved wiener filtering process to enhance its quality. Subsequently, Texton based features, Resnet based features, VGG 16 based features, Improved Semantic features and Improved LTP features are extracted from the preprocessed image. The last step in this model is future location prediction, a hybrid deep learning architecture is proposed in this step, which is the combination of improved LSTM and Bi-GRU model. Finally, Yolo V7 technique is used for the tracking purpose. Experiments on an available public dataset demonstrate that the proposed work is efficient than the conventional models on predicting the trajectories.

Improved filtering, Deep Learning, ILTP, Human Trajectory, Yolo V7