Abstract

Gesture recognition based on sign language is significant in forensic investigations and security surveillance, especially when monitoring non-verbal communication in sensitive settings is required. Nevertheless, current recognition systems often struggle with complex spatiotemporal dependencies and real-world variations, such as lighting, occlusion, and signer variation. To overcome these issues, this paper will develop a combined Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) system that leverages spatial-temporal features of sign language to analyze sign language gestures for forensic and security applications. The proposed methodology uses a deep CNN to extract high-level spatial features of video frames, including hand shape, orientation, and movement patterns. An LSTM network is then used to process these spatial features in sequence to model temporal dynamics in a gesture sequence. The framework has been tested on benchmark sign language video datasets containing more than 25,000 gesture samples across various sign categories. The performance measures were accuracy, precision, recall, and F1-score. The results of the experiment show that the CNN-LSTM model achieved higher overall recognition rates (94.8) than the traditional CNN-only and handcrafted feature-based models, by 8.6 and 14.2, respectively. The proposed system was also found to be more robust under noisy, low-resolution conditions, achieving an accuracy of 92.1 in simulated surveillance. Moreover, misclassification of visually similar gestures was reduced by 31 % through temporal modeling with LSTM. Finally, the integrated CNN-LSTM is a robust, scalable system for recognizing sign language gestures in law enforcement and security applications. Its capability to capture spatial and temporal properties enhances its interpretability and operational effectiveness, thereby boosting its application in automated surveillance, evidence analysis, and inclusive security communication systems.