Detection of salient objects is an emerging filed in computer vision, in which we usually detect prominent objects. We firstly perform object detection in the frame and then salient object is separated from the other objects. For the segmentation of salient objects we may require depth information of the detected objects. Many algorithms have been developed till now based on different approaches like YOLO, RCNN, SSD etc. YOLO particularly, uses a CNN for object detection and runs over the full image one time, dividing the image into small regions. It then makes the prediction on the generated bounding boxes for each region. The thesis aims in detecting prominent objects which humans can differentiate within a set of objects detected in an image. Due to the remarkable advancements in the fields of image recognition, localization and parallel processing power of GPUs, we will implement our system on GPU-based processing unit that is Jetson Nano and Kinect as our input image sensor.