Recognizing criminal faces from low-quality videos has been a persistent difficulty for law enforcement authorities. Conventional face recognition algorithms often fail to identify faces in low-resolution or low-light environments. This has led to a challenging task during criminal investigations. In re-cent years, deep learning-based algorithms have shown promise for increasing face recognition accuracy in demanding environments. Using GFPGAN, SURF, and HOG algorithms, this study offers a unique method for criminal face identification and recognition. GFPGAN is used to improve the quality of low-resolution pictures, while SURF and HOG are used to extract and match face characteristics. The proposed PATROL’S PAL architecture, with the above mentioned deep learning techniques was tested using a dataset of low-quality CCTV footage given by law enforcement officers of Thiruvannamalai District, India. The findings were helpful during their investigation procedure. The findings reveal the fact that there is a close similarity be-tween the images of, the accused individual caught during the ATM theft investigation, the suspect’s image from the database and the accused individual from the CCTV footage. Similarity between those images were obtained from the pro-posed architecture, with the threshold of Euclidean distance metric being set as 0.6. This highlights the efficacy of PATROL’S PAL in accurately identifying and confirming the accused individual. The enhanced accuracy of facial recognition in low-quality videos may help in narrowing down with the suspects quickly, thus by lowering the time and resources necessary for criminal investigations. The suggested method can be integrated with current surveillance systems, enabling real-time face identification under demanding situations.