I. INTRODUCTION:

One of the most exploited fields of computer vision and image processing is the human fall detection by means of surveillance camera. The main purpose of this is to help the old aged people, those under medical observation and kids if they fall and the intensity of injury is so much that they are not able to call someone to their rescue. In some cases, it may even avoid fatalities due to serious harms that may be caused due to the accidental fall. There are several methods such as manual help button that can be worn on hand but the main issue with this is that people may at times forget to wear such gadgets or after an intense, abrupt fall they may become benumb that they may not remember to press the help button; also in recent days sensor based devices such as accelerometer were used but they did not prove to be much efficient. Evolution in the field of computer vision has helped find an advanced method and hence the purpose is served by the intelligent surveillance camera.

Various semi-computerized video surveillance techniques are available that scrutinize the senior citizens, kids and people under medical observation to prevent fall accidents and provide medical treatment on time but this is quite a very tedious and inefficient task. Hence, this problem requires an intelligent solution wherein the entire video is not only captured but also used to monitor and infer falls which send in alarms to concerned people immediately.

II. RELATED WORK:

Recently researches have been done to analyze the posture of humans [1],[2],[3],[4],[5],[6],[7],[8] analysis about Human shape [9],[10],[11],[12],[13] and analysis of Human Motion [14],[15], and these factors are used to detect any chance of Fall occurrence. In these researches various methods are chosen in order to detect any unexpected change in the shape/posture/motion of the human taken care of. Likewise drawing ellipse around the shape of the body of human [1],[2],[3] or a box [10] or even a method in which orientation of body is determined by drawing line connecting three centroids expected to prevail in upper, middle and lower part of the body [9].

Applying above techniques many researchers have tried to develop an optimized fall detection technique. Nasution et al. [5] proposed an inventive method to detect fall by removing background from the image-frame captured in the video, and only shape of Human is left which passed to the k-NN classifier yields an accuracy of about 90%.

III. PROPOSED METHOD:

The method proposed to detect human fall and send in alarms for immediate response in emergency situations involves the use of various techniques where analysis of shape and position of a person in a frame of video is done, gradients in horizontal and vertical directions are calculated and changes in the time domain of the video are noticed and the system is trained with previous images dataset to improve the fall detection.

The proposed method suggests the installation of multiple cameras in order to provide the view of the entire indoor. This is so because if there is a single camera installed in a room then if the fall occurs in extreme proximity to the camera or if it is parallel to the camera, then the fall is not detected. So, multiple cameras resolve this issue as one camera will overcome the blind spots of other cameras. However, this requires a real-time syncing of the videos from multiple cameras into a single video frame for each time domain.

As a first step to this single generated video sequence, background subtraction is carried out in order to identify the foreground objects. It detects the mobile objects by making use of the distinction between the ongoing frame and the background model. For this purpose, the Gaussian Mixture Model as projected by Zivkovic[16] is used, in case when the background is of varied multitude, to further enhance the computational efficiency and performance of the system. To further remove the noise from these images, erosion and dilation techniques are deployed.

Thereafter, an edge-based detection technique in which contour based human template matching is applied to categorize whether it is human or non-human object. In this, the contour object from foreground frame is cropped to create a contour template and is compared with a predefined human template image to create a score. If this score corresponds to a defined threshold value, then the object in video frame is a human else it is not a human. A sudden abrupt change in human posture can also be recognized using this score. This is because when the human falls, there is an immediate decrease in this score which is persistent as long as the body comes into a static pose.

The human fall is determined by the separation between the middle center and top-center and the ratio of height to width of the rectangle around the human at the time the corresponding score gets abruptly changed. Generally, the height to width ratio for a human may vary from 3:1 to 4:1. The gap between the mid centre and top centre of rectangle is so used to determine the fall incident as it surpasses the situations when the person might be sitting straight or crouching down position as the difference between height and width of human in these situations is very small.

Fig 1: Flowchart of Proposed Method

The three parameters that are affected by the abrupt change in the human posture are: the matching score, the separation between top and mid center of rectangle surrounding the human body and the height-width ratio of the human. If there is a change in the score, then other two parameters are calculated and if they result to be less than the threshold value, then the human body is analyzed for 100 succeeding frames to check if it’s inoperative. Meanwhile, the system also analyses the situation using the dataset it has been trained with to confirm the fall chances. If the system ensures that the fall has occurred, then it sends in alarm to the close persons guaranteeing a human fall.

The dataset used to train the system must consists of huge amount of video sequences containing both images of fall(unusual human behavior frames wherein it involves forward, backward, sideways falls or falls due to the lack of balance) and non-fall (daily-routine human activities like sitting, walking, running, leaning and crouching down). These video frames must be at different positions and at different angles of the room. The falls must be recorded for both standing as well as sitting positions. The dataset must be catalogued for non-uniform illumination of the entire room. Also, the varying intensities of contrast between the subject’s clothing and the surrounding must also be taken care of while recording the dataset so as to yield better quality results. Different subjects with respect to gender, age, height and weight must participate in the dataset training samples. After the training examples have been collected, then Convulsion Neural Networks must be applied to it to train the system. Further, the dataset must be randomly divided into training and test set in the ratio of 80:20%. The feedback received from the system must be used further to enhance the accuracy of this model.

IV. CONCLUSION:

In this paper, the problem of developing a system for observation of human falls and generating alarms by keeping track of the videos of multiple cameras in indoors is considered. The dataset for a room is collected and the system is trained using CNN. Background Subtraction method is used for the extraction of objects in the foreground. The matching score, the separation between top and mid center of rectangle surrounding the human body and the height-width ratio of the human is used to detect any deformation in shape of human body and helps in detection of fall which could be done by keeping in track of human shape/posture/motion. This method can be improved with the use of cameras capable of taking snapshots from different angles at a time so that better calculation could be done required to detect Human Fall. The future scope of this paper deals with enriching the dataset, working on various other models to improve the accuracy of the system along with extending its scope to outdoor fall detection.

V. REFERENCES:

1. Nicolas Thome, Serge Miguet, and Sébastien Ambellouis “A Real-Time, Multiview Fall Detection System: A LHMM Based Approach”, IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 18, NO. 11, pp. 1522-1532, November, 2008.

2. Miao Yu, Adel Rhuma, Syed Mohsen Naqvi, Liang Wang, and Jonathon Chambers, “A Posture Recognitio Based Fall Detection System for Monitoring an Elderly Person in a Smart Home Environment”, IEEE Transactions ON Information Technology In Biomedicine, Vol. 16, No. 6, pp. 1274-1286, November 2012.

3. Homa Foroughi, Baharak Shakeri Aski, and Hamidreza Pourreza, “Intelligent Video Surveillance for Monitoring Fall Detection of Elderly in Home Environments”, Proceedings of 11th International Conference on Computer and Information Technology (ICCIT 2008), pp. 219-224, 25-27 December, 2008, Khulna, Bangladesh

4. Z. A. Khan, W. Sohn, “Abnormal Human Activity Recognition System Based on R-Transform and Kernel Discriminant Technique for Elderly Home Care,” IEEE Transactions on Consumer Electronics, Vol. 57, No. 4, November 2011.

5. Arie Hans Nasution and Sabu Emmanuel, “Intelligent Video Surveillance for Monitoring Elderly in Home Environments”, IEEE MMSP, pp. 203-206, 2007.

6. Chien-Liang Liu, Chia-Hoang Lee, Ping-Min Lin, “A fall detection system using k-nearest neighbor classifier”, Expert Systems with Applications, Elsevier, Vol. 37, pp. 7174-7181, 2010.

7. Chia-Feng Juang and Chia-Ming Chang, “Human Body Posture Classification by a Neural Fuzzy Network and Home Care System Application”, IEEE Transactions on Systems, Man, And Cybernatics— Part A: Systems and Humans, Vol. 37, No. 6, pp. 984-994, November 2007.

8. Damien Brulin, Yannick Benezeth, and Estelle Courtial, “Posture Recognition Based on Fuzzy Logic for Home Monitoring of the Elderly”, IEEE Transactions on Information Technology In Biomedicine, Vol. 16, No. 5, pp. 974-982, Sept. 2012.

9. Jia-Luen Chua, Yoong Choon Chang, Wee Keong Lim, A simple vision-based fall detection technique for indoor video surveillance”, Springer-Verlag, Vol. 9, pp.623-633, 2015.

10. Hong Liu, Changling Zuo, “An Improved Algorithm of Automatic Fall Detection”, AASRI Conference on Computational Intelligence and Bioinformatics, Elsevier, pp.353-358, 2012

11. Edouard Auvinet, Franck Multon, Alain Saint-Arnaud, Jacqueline Rousseau, and Jean Meunier, “Fall Detection with Multiple Cameras: An Occlusion-Resistant Method Based on 3-D Silhouette Vertical Distribution”, IEEE Transactions on Information Technology in Biomedicine, Vol. 15, No. 2, pp. 290-300, March 2011.

12. Homa Foroughi, Alireza Rezvanian, Amirhossien Paziraee, “Robust Fall Detection using Human Shape and Multi-class Support Vector Machine”, Sixth Indian Conference on Computer Vision, Graphics & Image Processing, IEEE, pp. 413-420, 2009.

13. Caroline Rougier, Jean Meunier, Alain St-Arnaud, and Jacqueline Rousseau, “Robust Video Surveillance for Fall Detection Based on Human Shape Deformation”, IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 21, NO. 5, pp. 611-622, MAY 2011.

14. Nicolas Thome and Serge Miguet, “A HHMM-Based Approach for Robust Fall Detection”, ICARCV, IEEE, (2006).

15. Damien Brulin, Yannick Benezeth, and Estelle Courtial Posture Recognition Based on Fuzzy Logic for Home Monitoring of the Elderly, IEEE Transactions on Information Technology in Biomedicine, Vol. 16, No. 5, (2012) pp. 974-982

16. Zoran Zivkovic, Ferdinand van der Heijden, “Efficient adaptive density estimation per image pixel for the task of background subtraction”, Pattern Recognition Letters, Vol. 27, pp. 773-780, 2006, 2006. doi:10. 1016/j.patrec.2005.11.00

17. Cohen, J, A coefficient of agreement for nominal scales, Educational and Psychological Measurement, 1960, 20 (1), pp. 37–46.

Received on 25.01.2020 Modified on 07.02.2020

Research J. Science and Tech. 2020; 12(1).65-68.

DOI: 10.5958/2349-2988.2020.00008.X