Pedestrian Presence Detection in Areas of Interest Using Multiple Cameras

Monitoring pedestrians in security scenarios are problems that cover different environments of society. Seeking solutions to this problem, the techniques of detection and tracking of pedestrians can meet the need for constant surveillance. In addition, many solutions involve long training and require much effort, increasing costs. To understand and find techniques that enable the detection of pedestrians in areas of interest, aiming at monitoring the safety of pedestrians, we carried out a comparative evaluation between a pedestrian detection technique and a pedestrian tracking technique, in addition to inserting the validation of pedestrian crossing in areas of interest in the image. We established complementary scenarios that seek a relationship between the pedestrian identification techniques in each scenario, obtaining average results regarding pedestrian detection inside the areas of interest. For the detection technique, we obtained an accuracy of 0.952 and an f-score of 0.970, and for the tracking technique an accuracy of 0.948 and an f-score of 0.967, with a precision of 0.977 and 0.984 for the detection and tracking techniques, respectively. For the number of frames with pedestrian presence in the areas of interest, we obtained the following average results for the pedestrian detection technique: accuracy and f-score of 0.985 and precision of 0.997. For the tracking technique, the accuracy was 0.973, and the f-score was 0.970, and this technique managed to obtain a precision of 1.000.