Development of Framework for Finding Missing Individual using Machine Learning

Abstract

The most critical aspect of global crisis of missing persons is time since the possibility of successful recovery is higher. reduces significantly during thefirst24to48hours. Sincetheissueislargeandurgent, conventionalmethods of investigation that are used are inadequate. mainly dependent on the manual examination of colossal quantities of visual proofs in the types of CCTV recordings andcitizenleads-areessentiallyinsufficient. Theseprocessesaretediousinnature, subject to humanerrorandexhaustion, and they lead to significant operation. bottlenecks stressing law enforcement forces and increasing the misery offamilies.Thisstudy directly deals with this. urgent requirement creating an integrated, machine learning-based pipeline in order to detect and identify missed persons. The proposed modelisnotjustameretool official recognition, butisamulti-algorithmsframeworkwhichisunified. works as a complete detection system. To start with, a detailed analysis of the face is conducted with the help of the mediapipe Face Mesh that is able to. detection of 468 3D face landmarks in real-time. To abstract a face image of a raw array of pixels into a structured/geometricimage.representation,itshouldbe capable of surviving the real world differences in pose, light and image quality. The core of our system uses a hybrid approach to feature extraction, in which geometric features based onfaciallandmarksare extracted using a bespoke, lightweight model. Convolutional Neural Network (CNN). The design allows acquisition of highly discriminative features of landmark spatial. relationships, creating compact, high-fidelity embedding vectors which in a unique way describe the facial geometry of an individual. This approach has high accuracy and computational efficiency needed in resource constrained environments. For identificationisamatchingenginebasedonK-NearestNeighbours(KNN)whichindexesallregistered embeddings into a feature which can be searched. space. When a new query is processed, a similarity search is conducted at a high speed with the help of the algorithm. An adjustable distance. threshold approves the possible matches, which guarantees high confidence and reduces false positives. An efficient and efficient identification. The output of the landmark detection, deep feature extraction, and efficient similarity search are a combination of workflow. This work provides an effective resource to bolstercommunityandlawenforcementworkbytransformingaconventionalsearchprocessintoa. Memberof the automated and data-driven procedure. It aims at speeding up the most important job of finding missing persons, reducing thetimeofinvestigations, and increase success rates.

Introduction

The text addresses the urgent global issue of missing persons, emphasizing that the first 48–72 hours are critical for recovery. Traditional investigation methods, such as manual analysis of CCTV footage, are slow, inefficient, and prone to human error, creating a need for automated, intelligent systems.

To address this, the study proposes a machine learning–based facial recognition system that automates identification using deep learning and geometric facial features. Unlike heavy models like FaceNet, the system balances accuracy and efficiency by combining MediaPipe Face Mesh (for extracting 468 facial landmarks) with a lightweight CNN to generate feature embeddings, and a K-Nearest Neighbors (KNN) algorithm with a dynamic threshold for accurate matching.

The methodology involves preprocessing images, extracting landmarks, generating embeddings, and performing fast similarity searches using Ball Tree indexing. The system is implemented using tools like TensorFlow, OpenCV, and Streamlit, enabling real-time, scalable, and low-cost deployment.

Results show that the system significantly reduces identification time from days to minutes, while maintaining high accuracy with low computational cost. Key advantages include efficiency, scalability, robustness to real-world variations, privacy preservation, and explainability.

However, limitations include reduced performance with heavy occlusion, lack of age progression handling, and reliance on frontal images. Future improvements include adding multimodal biometrics, GAN-based age adaptation, federated learning, and real-time CCTV integration.

Conclusion

This paper introduces a new hybrid machine learning pipeline to identify missing persons automatically, whichcan be effective in terms of accuracy, computational efficiency,andprivacyprotection.Thesystemcanbetrained with MediaPipe Face Mesh to extract geometric landmarks and a small CNN to generate discriminative embeddings in order to reach competitiveaccuracyandstillbedeployedon resource-constrained devices. The K-Nearest Neighbors matching engine is matched with Ball Tree indexing that guarantees identifying large databases in real time that are scalable. [14] The suggested framework addresses the most important constraints of the current solutions namely: the computational complexity of contemporary deep learning models and the ineffectiveness of traditional manual investigation methods. This system can significantly reduce the length of time usedininvestigations,increasetherateof recoveries and provide law enforcement agencies with a powerful tool to combat the problem of missing persons around the globe as byturningatime-sensitivesearchintoa data-driven and automated process.It is placed as a modular, morally responsible contribution to the applied machine learning discipline in the area of public safety due toitsexplainablegeometricrepresentationsand privacy-sensitive design. [4].

References

[1] J. Jacobo, \"Why the first 72 hours in a missing persons investigation are the most critical, according to criminology experts,\" ABC News, Oct. 2018. [2] E. Shaabani, H. Alvari, P. Shakarian, and J. E. K. Snyder, \"MIST: Missing PersonIntelligenceSynthesisToolkit,\"arXiv preprint arXiv:1607.08580, Aug. 2016. [3] \"Missing Person Detection,\" Journal of Emerging Technologies and Innovative Research (JETIR), 2024. [4] D.P.Shuklaetal.,\"ASurveyofIntegratingDeep Learning-Based Missing Person Detection Model into CCTV Systems for EnhancedIdentification,\"InternationalJournalof Advanced Research in Science, Communication and Technology(IJARSCT),vol.4,no.2,pp.632-637,Feb.2024. [5] R. Patel et al., \"Deep Learning for Understanding Faces,\" IEEE Signal Processing Magazine,JohnsHopkinsUniversity, 2018. [6] R. K. Sharma et al., \"Suhail: ADeepLearning-BasedSystem for Identifying Missing People,\" Computer and Information Science, vol. 16, no. 1, pp. 1-15, Apr. 2023. [7] F. Schroff, D. Karimov, and J. Philbin, \"FaceNet: A Unified Embedding for Face Recognition and Clustering,\" in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 815-823, 2015. [8] M. A. Khan et al., \"Lightweight Convolutional Neural NetworksforPersonRe-Identification,\"inProc. 2021 IEEE International Conference on Consumer Electronics (ICCE), June 2021. [9] S. Kumar et al., \"Person Re-identification UsingConvolutional Neural Networks,\" International Journal of Scientific and Research Publications(IJSART),vol.10,no.4, Apr. 2024. [10] A. Ahmed et al., \"A Novel Integration of Face-Recognition Algorithms with a Soft Voting Scheme for Efficiently Tracking Missing Person in Challenging Large-Gathering Scenarios,\" Sensors, vol. 22, no. 3, p. 1153, Jan. 2022. [11] S. Ahmad et al., \"Applying Machine Learning and DataFusion to the Missing Person Problem,\" TechRxiv Preprint, Sept. 2021. [12] K. Priya et al., \"Person Re-Identification in Criminal Investigations Using Deep Learning and Vision-Based Technique,\" in Proc. 2024 IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT), Dec. 2024. [13] V.Bazarevsky,Y.Kartynnik,A.Vakunov,K.Chesakov,and Y. Camporese, \"Real-time Facial Surface Geometry from Monocular VideoonMobileGPUs,\"inProc.2019IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2019. [14] M. Dolatshah, A. Hadian,andB.Minaei-Bidgoli,\"Ball*-tree: Efficient spatial indexing for constrained nearest-neighbor search in metric spaces,\" arXiv preprint arXiv:1511.00628, Nov. 2015. [15] H. Rehman, \"468-Face-Landmarks-of-Face-with-MediaPipe-Google\'s-Libr ary-in-Python-OpenCV,\" GitHub Repository, Feb. 2023. [16] X. Yu, J. Lin, J. Brandt, and Z. Lin, \"Pose-free faciallandmark fitting via optimized part mixture models,\"inProc. 2013 IEEE International Conference on Computer Vision (ICCV), 2013, pp. 1944-1951. [17] Y. Lipman, R. A. Al-Aifari, and I. Daubechies, \"The continuous Procrustes distance between two surfaces,\" arXiv preprint arXiv:1106.4588, June 2011. [18] K. Sokal, \"MediaPipe 3D Face Transform,\"Google Developers Blog, Sept. 2020. [19] T. Stojnic, \"Comparative Study of Relational and Document Databases,\" Journal of Database Systems, vol. 15, no. 2, pp. 45-62, 2023. [20] A. Khan, R. Sharma, and P. Singh, \"IdentificationofMissing Person using CNN,\" International Journal of Emerging Technologies and Innovative Research (IJISRT),vol.6,no.8,pp. 234-241, Aug. 2024. [21] N.Kumar,A.Belhumeur,andS.Nayar,\"Facerecognitionand the problem of aging,\" in Proc. 2016 IEEE/CVF Conference onComputerVision and Pattern Recognition (CVPR), 2016, pp. 1-8.

Copyright

Copyright © 2026 Aadi Yadav, Arpita Arora, Aarav Jain . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.