PERSONTRACK INTELLICAM AI based person detecting and tracking

PERSON-TRACK INTELLI-CAM AI based person detecting and tracking camera

ABOUT ME I am an individual innovator with specialization in Robotics and Artificial Intelligence/ Deep Learning.

PROBLEM STATEMENT a) Cross Border Infiltration Detection: Especially along the border which India shares with Pakistan and Bangladesh. The invention will help detect cross border terrorist movement as well as illegal migration and smuggling. b) Intrusion detection at high security places like Cantonments, Airports, Power plants, etc, where person entry is restricted. c) Automatic cameraman: Since the invention is also capable of tracking the detected person, it can be used in areas where a cameraman is required to follow a person, for example, recording lectures where the lecturer is moving, or stage performances or in sports or in movies. d) Disaster Relief: The camera can be mounted on a rover or a drone and can be used to automatically spot stranded people and deliver relief packages to them. e) Poacher detection in protected forests (as camera is able to differentiate

HOW DO THEY ADDRESS IT TODAY? -Sound based sensors: cannot differentiate between human and animal, also triggered by rainfall, high winds etc. -Motion detection cameras: they are based on background subtraction and hence accuracy is very low as they are triggered by false alarms like moving branches of trees, or movement of shadows during the day, etc. -PIR based sensors: they detect heat and hence cannot differentiate between humans and animals as well as range of operation is also very low (< 5 m). -Cloud based AI detection: Range of detection is low. Also, since data is sent to cloud, it is not safe and it is susceptible to data theft as well as hacking and Do. S attacks. Also, good internet connection is necessary and still there will be significant latency. -Satellite imagery/ drones: Need a human operator all the time and that operator can only scan a limited area at a time. -Non AI based cameras: like those performing skin-tone detection or body

INNOVATION The invention includes a camera which detects the location of the person in the video frame using Artificial Intelligence on device and is capable of tracking/ following the moving person in real time via a motor. Salient Features: • Person detection is performed using deep learning algorithm (AI) which has been trained to detect only persons and therefore accuracy of detection is better than other non AI based solutions and speed is better than other AI based solutions. • The camera is totally autonomous and does not require a human operator.

• Along, with detection, it can also track the person i. e. the camera follows the person (via motor on which the camera is mounted) as the person moves, hence not allowing the person to leave the Field of View of the camera. • It has night vision capability. • The range of detection is 2 m to 200 meters (Range can be increased as it depends on lens used). • Alongwith the detection and tracking, the system is capable of generating alerts via alarms, lights, SMS, Wifi, OFC based messages and transmit live video feed. PATENT DETAILS: Provisional patent application has been filed (dt 14/08/2019 ) with Application

Development Stage: A working prototype has been developed. 3 rd Party Verification: A novelty search was carried out for the sake of determining patentability of the invention by i. Cuerious Research Services LLP (an intellectual property research and technology consulting firm). Excerpt from the novelty search report : “the invention is different from the prior arts since the invention discloses the combination of an artificial intelligence algorithm and a tracking algorithm. ”

SOLUTION How the solution is better than other alternatives : • Person detection is performed using deep learning algorithm (AI) which has been trained to detect only persons and therefore accuracy of detection is better than other non AI based solutions. • It is capable of differentiating between a human and any other object like an animal, tree, car, chair, gate, door, etc. Therefore, it can be used to detect human presence in various environments both indoors and outdoors. • It is not based on motion detection or background subtraction. Therefore, it will not get triggered due to any motion like moving animals or insects, or opening/ closing of gates/doors/windows, change of lighting or shadows, or movement of branches/ leaves due to wind. Hence, limiting False Alarms. • Unlike PIR based sensors (which detect heat), its range of detection is larger and it also works outdoors with same accuracy as indoors.

• Person detection using AI is performed on device, therefore no internet connection (cloud) is required and data is not shared (privacy) with anyone. { Existing person detection cameras are connected to cloud servers (where the actual algorithm is running), the cameras do not have in built capability (they just relay the image to the servers). This increases latency and therefore makes real time tracking difficult, also internet connection is must and security issue is also there as images are sent to cloud. } • High Frame Rate of 11 FPS achieved on device which is sufficient to detect a running person. This high frame rate has been possible as the AI algorithm has been optimized and trained to detect only persons. • Night vision capability achieved by training the AI on IR images. • Range of detection: 2 metres to 200 metres (range depends on lens used and therefore can be increased). • Live tracking feature i. e. the camera is mounted on a motor which follows the detected person, therefore not allowing the person to leave the Fo. V of the camera.

• Low Power requirements : 5 V/2. 4 A and therefore can be powered even via a phone battery bank for upto 12 hours. • Small Size : 6 cm x 6 cm x 15 cm

USE CASES a) Cross Border Infiltration Detection: Especially along the border which India shares with Pakistan and Bangladesh. The invention will help detect cross border terrorist movement as well as illegal migration and smuggling. b) Intrusion detection at high security places like Cantonments, Airports, Power plants, etc, where person entry is restricted. c) Automatic cameraman: Since the invention is also capable of tracking the detected person, it can be used in areas where a cameraman is required to follow a person, for example, recording lectures where the lecturer is moving, or stage performances or in sports or in movies. Eg: the output from the camera can be used to rotate the motor on which a high-definition slowmotion close-up camera is mounted. This will enable the slow-mo camera to follow the moving person automatically, especially in individual sports like tennis and badminton.

d) Automatic sentry gun: The output from camera can be used to control a gun which can automatically aim and shoot at moving intruders. e) Disaster Relief: The camera can be mounted on a rover or a drone and can be used to automatically spot stranded people and deliver relief packages to them f) Poacher detection in protected forests (as camera is able to differentiate between animals and humans).

INTRUSION DETECTION From Far Away High Location

Intrusion Detection in difficult environments Difficult background with trees, bushes, benches etc, and low light conditions Camouflaged Uniform Night Vision Capable

AUTOMATIC CAMERAMAN

WORKING OF AUTOMATIC CAMERAMAN Intelli. Cam Slow Motion HD Camer being controlled by Intellicam via a moto

DISASTER RELIEF Detection of people stranded in disasters.

MARKET SIZE Target customers include : Ø Companies dealing with security and surveillance solutions. Ø Defence research organizations like DRDO. Ø Government : Border Security, Disaster Relief, Anti Poaching, Homeland Security, Education (automatic recording of lectures for online courses). Ø Companies dealing with video recording entertainment industry especially sports, theatre, movies. Eg: the output from the camera can be used to rotate the motor on which a high-definition slow-motion close-up camera is mounted. This will enable the slow-mo camera to follow the moving person automatically, especially in individual sports like tennis and badminton.

Border Security: • Length of border India shares with Pakistan (3, 323 Km), Bangladesh (4, 096 Km), Myanmar (1, 643 km), China (3, 488 km), Nepal (1, 751 km), Bhutan (699 km) =15, 000 Km • Then out of this 15, 000 Km length of border, assuming 50% is to be put under this intrusion detection system of the cameras, we get 7, 500 Km long distance to be covered. • Taking, one camera unit has a range of intrusion detection of 500 m radius. Camera units required to cover this distance : 7500 km/1 km (500 m + 500 m) = 7500 • Hardware cost of one unit = Rs 5000 (approx) • Setting selling price to even Rs 25, 000 will make it cheaper than existing systems which include laser fencing (high per km cost), manual thermal cameras (each costing more than Rs 50 k), manual patrolling etc. • Therefore, market value for border security turns out to be = 7500 x Rs 20, 000 = Rs 15 crores

Similar estimation can be extended for Homeland Security which includes airports, power plants, cantonments etc. Taking example of (airports + cantonments + nuclear reactors), where average fencing length to be covered under intrusion detection system per airport, cantonment, nuclear reactor can be approximated to be 5 Km, 10 Km, respectively Total no. of units in India = 22 Nuclear reactors + 134 Airports + 62 Cantonments Total distance to be covered = 22 x 10 km + 134 x 5 Km + 62 x 10 Km = 1, 510 Km Therefore, minimum market value in just these 3 sectors = 1, 510 x Rs 20, 000 = Rs 3 crores Anti – poaching: With 105 national parks and 543 wildlife sanctuaries covering 1, 56, 719 sq km, approx fencing length comes out to be 50 km per national park/ sanctuary, therefore total length to be covered = 50 x (543 + 105) = 32, 400 Km.

Entertainment Industry: Huge market potential in individual sports like lawn tennis, badminton, table tennis, diving etc where it can replace cameramen controlling the High-def Slow-mo Close-up cameras. Also, in movies, the output from the camera can be used to control the closeup camera focusing on the actor, while the actor moves. Due to this huge potential, the estimated market value in the entertainment industry is easily more than Rs. 50 crores.

COMPETITION Security: LASTEC (Laser Fence system): Already deployed on Indo – Bangladesh border. Shortcoming : Laser fence is deployed along the fence and hence is easily detectable. Secondly, it is not able to differentiate between which object has intercepted the laser and set off the alarm. Therefore, many false alarms due to animals, leaves etc. Thirdly, cost and power requirement is more. Their competitive advantage : Already deployed and tested.

BUSINESS MODEL The Business Model is mainly B 2 B and B 2 G centric where the intention is to license the technology to companies and government for the aforementioned market.

PROFILE Graduation : B. Tech (2016) in Electronics and Communication from PEC University of Technology (formerly Punjab Engineering College), Chandigarh, India. COURSES: • Five Course Online Specialization in Deep Learning (Coursera) including Convolution Neural Networks, Recurrent Neural Networks, LSTM, etc. offered by Deeplearning. ai. • Online course on Machine Learning offered by Stanford University on Coursera. INTERNSHIP:

PROJECTS: • AI based Person following robot. • Face Recognition Camera onboard Raspberry Pi 3. • Human like robotic arm capable of copying human arm movement shown to it live using Kinect camera and capable of remembering an action taught to it. • Fall detection for elderly (wearable sensor using MPU 6050 accelerometer and gyroscope) capable of sending alerts using Bluetooth and calling relatives/emergency number. • HID (Human Interface Device) device for computers and PID controlled robots achieved through sensor fusion of accelerometer and gyroscope: A Multipurpose (Wireless+USB) HID Device capable of controlling the mouse cursor and keyboard of the computer through simple hand gestures. It has been optimized to work as : i) Air Mouse and a ball point pen. ii) Gamepad compatible with latest PC games providing life like control to gamers. • PID controlled Self Balancing Robot and Self Balancing Algorithm for Quadcopters.

• Lab. VIEW based home automation and security system: Which included Speech controlled light and fan, IR sensor for burglar detection, Ultrasonic sensor for water level indication, Fire detector and automatic water spraying mechanism, automatic solar panel positioning for maximum exposure, automatic image capturing upon doorbell and burglar alert, email alert to owner upon fire, doorbell and burglar detection along with attached image, and buzzer alerts. • Automatic traffic light control system: Depending on the volume of traffic, the duration of the red and green light is changed so as to ensure a smooth movement of traffic.