These days, deep learning and convolutional neural networks (CNN) are gaining popularity in terms performing far better compared to traditional methods when it comes to object detection, localization and classification, making the job easier for humankind. However, deploying them successfully to perform some certain tasks still remains to be an issue. For instance, in object detection and localization in a video or still image, since existing datasets include only limited number of class objects, sometimes object that needs to be detected and localized may not exist in the dataset, making it impossible to classify the object. Thus, to overcome this isssue, a new aproach has been proposed which makes use of the CNN-based YOLO algorithm. In this work, a new model has been developed which is able to classify and localize objects which don't exist in the dataset. With the help of this model, just like any other objects, unknown objects can be identified as well after sufficient training of the model. This trained model achieved high accuracy in terms of predicting objects right, making it reliable for other projects which involves detecting objects which are specific to new classes.