Object Detection Based on Deep Convolutional Neural Network
In the task of image object detection, a large number of candidate boxes are generated to match the small number of candidates containing images of object, which will increase the amount of data in neural networks and reduce the real-time of object detection. So, candidate boxes should be reduced in the training and evaluation of the network. In the other hand, hand-made data-set makes the object detection inefficient in the application scenarios where the recognition tasks are frequently changed. Here, we propose a new method about extraction of candidate boxes for object detection, and the automatic image annotation for data-set making. This method can be used in whose object always change, such as, Industrial assembly line, automatic recognition and counting of cells under a microscope and so on. The experimental data set shows that compared with the classical model, the new method can significantly improve the efficiency and real-time performance of object detection.