Adaptive behaviour tuning of a neural network-based method for moving object recognition in video streams

B. Sadovnykov; V. Lysechko

doi:10.36910/6775-2524-0560-2025-60-05

B. Sadovnykov https://orcid.org/0009-0009-4180-2863
V. Lysechko https://orcid.org/0000-0002-1520-9515

DOI: https://doi.org/10.36910/6775-2524-0560-2025-60-05

Keywords: video stream, computer vision, image processing, computational resources, neural networks, object recognition, adaptive algorithms, telecommunication systems

Abstract

The article presents an improvement of the method for searching and recognizing moving objects in video streams in real time, which is based on calculating interframe differences (deltas) and using a neural classifier. A mechanism for adaptive behaviour tuning of the method depending on the characteristics of the input data is proposed, which makes it possible to increase the recognition accuracy and processing speed under changing background conditions and limited computational resources. The developed method is an evolutionary adaptive mechanism, that allows the algorithm to gradually change its processing strategies based on the collected data, forming a heat map and optimising its performance for the specifics of a particular environment. To evaluate the effectiveness, an experimental comparison of the improved method with its basic version [5] was carried out, analyzing indicators such as average frame processing time, RAM and video memory usage, CPU and GPU load, and recognition accuracy. The optimization resulted in up to a 20% increase in processing speed and a slight improvement in accuracy (~0.8%) without increasing the use of key computational resources. The experimental results confirm the feasibility of integrating the adaptation mechanism into the delta-classification method to improve its efficiency for real-time operation.

References

1. Abdallah W., Val T. (2020). Genetic-Voronoi algorithm for coverage of IoT data collection networks. 30th International Conference on Computer Theory and Applications (ICCTA), Alexandria, Egypt, 2020. РР. 16-22.
2. Bernardin R., Paias A. (2018). Metaheuristics based on decision hierarchies for the traveling purchaser problem. International Transactions in Operational Research, 25(4), 2018. PP. 1269-1295.
3. Blank J., Deb K. (2020). Pymoo: Multi-Objective Optimization. IEEE Access, Vol. 8, 2020. PP. 89497-89509.
4. Blank J., Deb K., Dhebar Y., Bandaru S., SeadaH. (2012.) Generating well-spaced points on a unit simplex for evolutionary many-objective optimization. IEEE Transactions on Evolutionary Computation, 25(1), 2021. PP. 48-60.
5. Cheng R., Jin Y., Olhofer M., Sendhoff B. (2016). A reference vector guided evolutionary algorithm for many-objective optimization. IEEE Transactions on Evolutionary Computation, 20(5), 2016. PP. 73-791.