Multiple object detection and recognition from live video cameras using deep convolutional networks for autonomous vehicles

Abstract

In self-driving technologies the system perceived the environment without human intervention in which system can detect different obstacles and make decisions for smart transportation. In this studies we adapt and design different technique for detecting and recognize an objects which used different components like data processing, noise removal, input resize, input vector preparation, feature extraction, classification and regression problems. However, in the current research, the performance of the detector is not reach matured. And they used fully connected layers in detection networks for the detector models. Due to this problem the performance in the detection networks is not satisfied and doesn't extract many features. We developed a new model in detection networks using convolutional neural networks and extracted different level of features which helps the model to extract more usable information to the classification and regression problems in the detector. In the proposed model we used 3 layers of fully convolutional neural networks and two fully connected layers to develop the model. In the experiment, we have evaluate both the localization and classification mAP of the networks. And, we obtained 84% mAP model performance. And also, we evaluate the quantitative and qualitative results for the networks for each categories in the input data. Thus, our model, we detected more objects that doesn't detect in the previous works.