CLGA-Net: Holistic Hybrid Deep Learning and Optimization Framework for Advanced Motion Prediction and Analysis in Video Processing
Salini Abraham, Thirunavukkarasu Kanimozhi · 2025
For applications like surveillance, human activity recognition, and autonomous systems, the process of motion prediction from video processing is vital where traditional approaches suffer with factors like computational efficiency, accuracy, adaptability, and so on. Integrating deep learning architectures, a holistic hybrid DL and optimization framework is designed in this paper that is used in enhancing motion prediction. Infused with CNN for spatial features extraction and LSTM module for motion sequence learning, the structured holistic framework ensures accurate motion prediction. The CLGA-Net is further optimized using the hybrid optimization module that consists of Genetic Algorithm and Adaptive Gradient-based Optimization for hyperparameter optimization and fine-tuning weights. Through the validation on metrics like accuracy (90%), robustness, and efficiency of the model on the UCF101 dataset, the findings reveal that the proposed model is well suited for implementations like object tracking, anomaly detection, and gesture recognition.