Research has shown that there is an increase in number of elderly people living alone. Taking this into consideration, a smart detection system for elderly people has been proposed in the current research paper to detect fall of elderly people so as to prevent any serious injuries to the elderly people who stay at home by themselves. Fall has been identified as one of the leading causes of serious injuries and even fatalities. This system can assist caretakers and family members in monitoring elderly individuals, enabling them to take timely actions in any signs of a fall being detected. An alert notification is sent to the caretaker to provide medical care as early as possible to avoid any serious injuries. The work presented in the paper involves an approach to process the images and videos isolating the background and foreground and using optical flow and head speed movement for enhancing the accuracy of detection. Four deep learning models-VGG16, ResNet50, InceptionV3 and EfficientNetB7 were used for implementing the detection of falls. Two types of images/frames were considered, one with proper illumination and another with low light from www.falldataset.com and a created dataset. 1050 images with proper illumination and low light were used for training out of them, 150 images with proper light and 95 low light were used for testing. In both the categories of images, VGG16 accuracies reported 96% for images with proper illumination and 92.63% low light images while EfficientNetB7 reported 97.33% and 93.68% respectively. In both the categories VGG16 and EfficientNetB7 have outperformed better than the other two in terms accuracies. Also successful alert notification was demonstrated via email and phone number with hardware platform using Arduino Uno and GSM module for sending alert messages to registered phone number and Raspberry Pi 3 for sending to the registered email Id.