In this article we consider the application of the modern object detection method – Single Shot Multibox Detector. We have trained the convolutional neural network for vehicle detection on a sample of 3000 images with marked areas where are the vehicles are placed. A network quality check was performed on 7000 test images. The test and training samples contain images made by a monocular camera mounted in a vehicle moving along suburban highways during daylight hours. Recall and precision of object detection on the test sample is correspondingly more than 88% and 78%. Recognition of one frame takes 28.5 milliseconds. Experiment was performed on a graphics processor using NVidia CUDA technology. The obtained results can be applied in driver assistance systems and monitoring of the traffic situations.
In this article we consider the application of the modern object detection method – Single Shot Multibox Detector. We have trained the convolutional neural network for vehicle detection on a sample of 3000 images with marked areas where are the vehicles are placed. A network quality check was performed on 7000 test images. The test and training samples contain images made by a monocular camera mounted in a vehicle moving along suburban highways during daylight hours. Recall and precision of object detection on the test sample is correspondingly more than 88% and 78%. Recognition of one frame takes 28.5 milliseconds. Experiment was performed on a graphics processor using NVidia CUDA technology. The obtained results can be applied in driver assistance systems and monitoring of the traffic situations.