Background subtraction and yolo algorithm

Abstract

Introduction: This article is the result of research entitled “Signal processing system for the detection of people in agglomerations in areas of public space in the city of Cúcuta”, developed at the Universidad Francisco de Paula Santander in 2020.Problem: The high percentage of false positives and false negatives in people detection processes makes decision making in video surveillance, tracking and tracing applications complex. Objective: To determine which technique for the detection of people presents better results in terms of respon-se time and detection hits.Methodology: Two techniques for the detection of people in uncontrolled environments are validated in Python with videos taken inside the Universidad Francisco de Paula Santander: Background subtraction and the YOLO algorithm.Results: With the background subtraction technique, we obtained a hit rate of 84.07 % and an average response time of 0.815 seconds. Likewise, with the YOLO algorithm the hit rate and average response time are 90% and 4.59 seconds respectively.Conclusion: It is possible to infer the use of the background subtraction technique in hardware tools such as the Pi 3B+ Raspberry board for processes in which the analysis of information in real time is prioritized, while the YOLO algorithm presents the characteristics required in the processes in which the information is analyzed after the acquisition of the image.Originality: Through this research, aspects required for the real-time analysis of information obtained in pro-cesses of people detection in uncontrolled environments were analyzed. Limitations: The analyzed videos were taken only at the Universidad Francisco de Paula Santander. Also, the Raspberry Pi 3B+ board overheats when processing the video images, due to the full resource requirement of the device.