Object recognition in videos using Optical Flow and Gaussian Mixture Model

Yıl 2025, Cilt: 8 Sayı: 1, 23 - 32, 15.06.2025

Hozaifah Alkhattab , Fatih Bayram

https://doi.org/10.53448/akuumubd.1683501

Öz

This study proposes an approach that utilizes Optical Flow and Gaussian Mixture Model (GMM) for object recognition in videos. The aim is to enhance detection performance by perceiving human movements. Experiments were conducted using two different methods: (1) object recognition with background subtraction and (2) direct object recognition without background subtraction. The parameters measured in the experiments were processing time and recognition accuracy. The results indicate that, in general, detection accuracy is higher when background subtraction is not applied (for example, in the “Daria_walk” video, 84.47% accuracy with a processing time of 3.2553 seconds). However, the background subtraction method reduces total processing time (e.g., in the “Daria_walk” video, 80.23% accuracy and 2.999 seconds processing time). Similarly, in the “Daria_jump” video, background subtraction yielded 67.88% accuracy and 2.6983 seconds processing time, while without background subtraction, 83.83% accuracy and 2.811 seconds processing time were achieved. Both methods successfully detected the human object in all videos. This research contributes to human activity recognition methods in the field of computer vision and demonstrates its applicability in areas such as security and video surveillance.

Anahtar Kelimeler

Optical Flow, Gaussian Mixture Model, Object Recognition, Background Subtraction

Videolarda Optik Akış ve Gauss Karışım Modeli Kullanarak Nesne Tanıma

Öz

Bu çalışma, videolarda nesne tanıma için Optik Akış (Optical Flow) ve Gauss Karışım Modeli (GMM) kullanan bir yaklaşım önermektedir. Amaç, insan hareketlerini algılayarak tespit performansını artırmaktır. İki farklı yöntem ışığında deneyler gerçekleştirilmiştir: (1) Arka plan çıkarma ile nesne tanıma ve (2) arka plan çıkarılmadan doğrudan nesne tanıma. Deneylerde ölçülen parametreler, zaman performansı ve tanıma hassasiyetidir. Sonuçlar, arka plan çıkarma uygulanmadığında algılama doğruluğunun genel olarak daha yüksek olduğunu (örneğin, “Daria_walk” videosunda %84.47 doğruluk, işlem süresi 3.2553 saniye) ancak arka plan çıkarma yönteminin toplam işlem sürelerini azalttığını (örneğin, “Daria_walk” videosunda %80.23 doğruluk ve 2.999 saniye işlem süresi) göstermektedir. Benzer şekilde, “Daria_jump” videosunda arka plan çıkarma yöntemiyle %67.88 doğruluk ve 2.6983 saniye işlem süresi elde edilirken, arka plan çıkarılmadan %83.83 doğruluk ve 2.811 saniye işlem süresi elde edilmiştir. Her iki yöntem de tüm videolardaki insan nesnesini doğru şekilde tespit etmiştir. Bu araştırma, bilgisayarla görü alanındaki insan aktivitesi tanıma yöntemlerine katkı sağlamakta ve güvenlik, video gözetimi gibi uygulamalarda kullanılabilirliğini göstermektedir.

Anahtar Kelimeler

Optik Akış, Gauss Karışım Modeli, Nesne Tanıma, Arkaplan Çıkarma

Kaynakça

• Ali, S.T., Goyal, K. ve Singhai, J., 2017. Moving Object Detection Using Self-Adaptive Gaussian Mixture Model for Real-Time Applications. Proceedings of the International Conference on Signal Processing and Embedded Systems (RISE), IEEE.
• Amjad, M., Khan, A., Siddiqi, M.H., Lee, S. ve Muhammad, N., 2021. Hierarchical human activity recognition using wearable sensors based on subspace pooling. Sensors, 21(7), 2368.
• Attal, F., Mohammed, S., Dedabrishvili, M., Chamroukhi, F., Oukhellou, L. ve Amirat, Y., 2015. Physical human activity recognition using wearable sensors. Sensors, 15(12), 31314-31338.
• Chauhan, A.K., Krishan, P. ve Kumar, D., 2013. Moving Object Tracking Using Gaussian Mixture Model and Optical Flow. International Journal of Advanced Research in Computer Science and Software Engineering (IJARCSSE), 3(4), 243–246.
• Choi, H., Kang, B. ve Kim, D., 2022. Moving Object Tracking Based on Sparse Optical Flow with Moving Window and Target Estimator. Sensors, 22(8).
• Horn, B.K.P. and Schunck, B.G., 1981. Determining optical flow. Artificial Intelligence, 17, 1–3, 185–203. Jain, R., Kasturi, R. ve Schunck, B.G., 1995. Machine Vision. McGraw-Hill.
• Jiang, Y., Wang, J., Liang, Y. ve Xia, J., 2018. Combining Static and Dynamic Features for Real-Time Moving Pedestrian Detection. Multimedia Tools and Applications, Springer.
• Khan, N., Mehmood, W., Iqbal, T., Ahmad, S., Rehman, A. ve Rho, S., 2024. Multimodal human activity recognition using hybrid CNN-LSTM network. Heliyon, 10(4), e22371.
• Lucas, B.D. and Kanade, T., 1981. An iterative image registration technique with an application to stereo vision. Proceedings of the 7th International Joint Conference on Artificial Intelligence (IJCAI), 2, 674–679.
• Meghana, R.K., Chitkara, Y., Apoorva, S. ve Mohana, 2019. Background-Modelling Techniques for Foreground Detection and Tracking Using Gaussian Mixture Model. Proceedings of the Third International Conference on Computing Methodologies and Communication (ICCMC), IEEE.
• Mohana ve Aradhya, H.V.R., 2020. Performance Evaluation of Background Modeling Methods for Object Detection and Tracking. Proceedings of the Fourth International Conference on Inventive Systems and Control (ICISC 2020), IEEE.
• Pushpalatha, K. ve Math, S., 2022. Hybrid deep learning model for human activity recognition using convolutional neural network and gated recurrent unit. International Journal of Intelligent Technologies and Applied Statistics, 15(1), 1-13.
• Redmon, J. ve Farhadi, A., 2017. YOLO9000: Better, Faster, Stronger. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1, 6517–6525.
• Šilar, Z. ve Dobrovolný, M., 2013. The Obstacle Detection on the Railway Crossing Based on Optical Flow and Clustering. Proceedings of the Telecommunications and Signal Processing Conference (TSP), IEEE.
• Stauffer, C. and Grimson, W.E.L., 1999. Adaptive background mixture models for real-time tracking. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), 2, 246–252.
• Tao, D., Li, X., Wu, X. ve Maybank, S.J., 2017. Human action recognition based on multiple features using Gaussian mixture models. IEEE Transactions on Image Processing, 26(3), 1377-1390.
• Thotapalli, P.K., Kumar, C.R.V. ve Reddy, B.C.M., 2021. Feature Extraction of Moving Objects Using Background Subtraction Technique for Robotic Applications. International Journal of Intelligent Robotics and Applications (IJIRA), Springer.
• Yi, Q., 2018. Design of Moving Object Detection System Based on FPGA. Proceedings of the 10th International Conference on Communications, Circuits, and Systems (ICCCS), IEEE.
• Zebhi, B., 2022. 2D discrete wavelet and Fourier transform-based convolutional neural networks for human activity recognition. Computers in Biology and Medicine, 141, 105072.
• Zhao, X. ve Su, Y., 2017. A dynamic approach for background modeling using Gaussian mixture models. Computer Vision and Image Understanding, 157, 36-49.
• Zhou, D. ve Zhang, H., 2017. Modified GMM Background Modeling and Optical Flow for Detection of Moving Objects. Proceedings of the National Conference on Electronics and Computer Systems (NECS), IEEE.