With the development of the society and medical technologies, the average age of population keeps increasing, health care. I has been a worldwide problem need to be solved. Though hospitals can heal many patients send to them, falls are in fact the first rated incentive that causes disabilities among the elderly people due to long-time unnoticed can lead to serious sequences especially for the elderly and patients who live alone. Fall detection would play a key role in preventing elderly people and patients harm from longtime lying without noticed. The most important part of a fall detection algorithm is to distinguish fall from the other daily fall-like activities such as lie down and squat. This paper studied the algorithm of fall detection based on vision-approach, compared different approaches to detect falls and presents a new algorithm based on vision using multi-parameters to discriminate falls and other activities which turns out have a higher accuracy on fall detecting than some other vision-based fall down detection algorithms and the experimental results reveal this method made the fall detection more reliable.