Pendeteksian Objek Dalam Jaringan Sensor Nirkabel Multimedia

Abstract

Object detection applications can be used to perform surveillance functions on car traffic. A car can be identified by its type and license number (plate). Meanwhile, the reliability of multimedia sensor networks for car traffic monitoring can be considered based on the number of multimedia sensors used and the positioning of the sensors. This study uses a camera sensor with Phase Detection Auto Focus (PDAF) technology, similar to a pair of right and left eyes when looking at an object. This research discusses car detection by multimedia sensors in a wireless multimedia sensor network used for car traffic monitoring. The study used 2 car objects to detect the type and license plate. However, the research focuses on detection by proposing a car and license plate detection model called Faster RCNN IFY, where its detection performance is compared to the YOLOv5 model and Commercial ALPR applications. Based on the detection, the disparity value obtained at a maximum distance of 50 m is 6.20 x 106 pix/mm, while the depth image value at the maximum disparity value is 16.88 x 109 mm3/pix. Then, in the proposed model, with a testing dataset of 40 images using the Faster R-CNN IFY model, the average value of car plate detection accuracy is quite good, namely 80.67%. Although these results are not as accurate as when the same testing dataset is applied to the Commercial ALPR application, which is 87.9%. The more training datasets, the better the accuracy of the detection results. Therefore, in the Faster R-CNN IFY model with a training dataset of 240 images, the average accuracy of the car detection results is only 52.6%. Whereas in the YOLOv5 model, with a total training dataset of 1500 images, an increase in the average value of the accuracy of the detection results was obtained to be 87.1%. In the end, in this study the Faster R-CNN IFY model was superior in terms of speed compared to the Commercial ALPR and YOLOv5 models. The average time needed to test 40 car images using the YOLOv5 model is 111.43 msec. Meanwhile, applying the Faster R-CNN IFY model only takes 11.86 msec. Likewise, when detecting 40 car plate images using Commercial ALPR, it takes an average time of 288.76 msec, while using the Faster R-CNN IFY model it only takes 11.86 msec. In addition, this model is also able to simultaneously detect 2 object classes in a car, namely the car class and license plate class. Meanwhile, Commercial ALPR and YOLOv5 can only detect 1 class of objects in a car.