Regresar

Object detection in rural roads using Tensorflow API

Abstract:

Object detection is a challenge in the computer vision area. Traditional techniques work reasonably well for this problem in urban areas where the roads and the boundaries are clearly marked. However, in the rural areas in the developed countries, the assumption does not hold which leads to the failure of such techniques. In this paper, we propose using TensorFlow Object Detection API, based on the combination of deep convolutional neural networks (CNNs). We investigate the performance of Faster RCCN, RFCN, and SDD frameworks in the context of cars, people, bike, and motorcycle detection from rural area images. We trained and tested these models on our own dataset. The Faster RCNN Resnet 50 model provides high performance in the bike, motorcycle and, car recognition, however, the Faster RCNN inception v2 models are better at detecting bike and, motorcycle in complex scenarios.