| Object detection is one of the most classical computer vision task that is used to detect objects from an image.Efficient and accurate object detection plays a vital role in the advancement of computer vision that makes it able to detect the objects in real time applications with more accuracy and less processing time.With recent advancement in machine learning and deep learning techniques,the accuracy for object detection has significantly increased.Existing Convolutional Neural Network(CNN)based object detection techniques achieve much better accuracy than classical detection algorithms but it suffers from more training and testing time that limits the overall performance of system and make it impractical for real time applications.In order to deal with the aforementioned issues,we have proposed two techniques that not only increase the accuracy of system but also reduce the training and testing time which makes it practical for real time object detection applications.Firstly,we have created two indoor datasets which have never been created by anyone before and then data augmentation is applied to one indoor dataset which increase the data size that results better accuracy and avoids the issue of over-fitting.Further,we do transfer learning in Faster R-CNN and SSD techniques from layer 7 by setting different learning rates and weights along with different mini batch size of 64,128 and so on.In addition,we have used data reduction techniques to our proposed methods.Due to this,our proposed methods take less training and testing time(with very slight reduction in accuracy)while consume less memory and power that makes it more efficient than existing Faster R-CNN and SSD techniques.Simulation results validates that our proposed methods achieve better accuracy and take less training and testing time than existing techniques.Thus,it can be concluded that our proposed framework is applicable for real time indoor object detection. |
