Natural disasters like floods often occur due to heavy rainfall, storms, melting snow and ice, overflowing rivers, dam failures, and urban drainage systems. Failing to evacuate flooded watery areas leads to drowning of objects into the water. While drowning in water people feel difficult to breathe and may not survive for a long period. Real-time Detection of persons and vehicles in heavy water flow is a challenging task. This paper proposes a framework to identify floating and almost drowned objects in the water and classify them whether they are humans or non-living objects using a series of Convolution Neural Network object detection models. Faster RCNN, Mask RCNN and, You Only Look Once (YOLOv5) network models are trained on the image dataset. In case of overlapped objects, object segmentation is also performed using Mask RCNN for predicting the shape of the drowning object. Faster RCNN and YOLOv5 models are validated using a test dataset and a decline in training loss is plotted on a tensor board. Results for evaluating object detection model show Faster RCNN as the best model for detecting and classifying objects in water than YOLOv5 and Mask RCNN. Distance between object is measured to find the shortest path to reach the object for faster rescue operations. Counting and Tracking of objects are performed to know the exact count of objects who need help in an emergency and to monitor their real-time position in the water.