Video security incidents

vumacam
by vumatel isp, Community active protection CAP, jhb non profit company gets video material. Head is Sean Jammy. Rob Nichol head of AI Surveillance, Randburg.

security vehicle
http://www.news24.com/Video/SouthAfrica/News/watch-robbers-rammed-with-cash-in-transit-vehicle-20171012. A Yolo controlled Gimbal with a PepperBall gun is pushed up through the roof opening in the corner of the cash in transit vehicle with  Linear actuators. Actuation of such a device must always be done physically, some physical lever must be pulled to prevent the AI system from auto engaging. Fit a Quad frames drone with a Flame thrower. Control engagement with an acoustic gunshot detecor and AI network, this is to prevent an aggressive criminal prosecution as SA law dictates that your life must be in immediate danger for lethal force to be used. Having a drone fitted flame thrower had better be accompanied by video proof that no other reasonable option was available(be certain to confirm this usage with the https://www.npa.gov.za/node/4 themselves). Create various simulated armed assault, counter flame throwing defence engagements with the Unreal graphics engine and provide the youtube github link so that the NPA can comment on it.(Think of South Africa as being in a Buck Rogers adventure.)

Car jacking
https://www.youtube.com/watch?v=SJDO_QjAT70 Every 32 minutes, a motorist is hijacked somewhere in South Africa - that’s a horrifying 46 cars hijacked every day. Hijacking and car theft are, unfortunately, a reality which have to be factored into our daily lives. We’re faced with decisions like whether or not to get insurance, install a tracker or park our car on the verge at our best friend’s braai.

jumping out of car
https://www.youtube.com/watch?time_continue=1&v=SHi9MuW1wSs four occupants disengage from vehicle after stopping in the road.

Remote gate opening
'''Install Dragon teeth at security booms, opening and closing it by remote activation and monitoring. The person who opens the gate must not be in a position where he could be killed for refusing to do so. '''

Install dragon teeth below a public road surface covered by as metal plate allowing for unimpeded traffic and remote activation only on an incident.

In Gauteng recently there was an incident where a screened worker passing all the checks at the entrance walked up to a parked security car and pushed a gun against the guards head, ordering him to drive to the exit. At the exit he threatened the unarmed guards to open the gate, else he would kill the security officer he is kidnapping behind the wheel.

What they should have done was to place a board outside explaining that the guards at the gate whom can be attacked have no power over the dragon teeth, only remotely over the Internet from a secured office can it be opened. The remote guard speaks to the would be kidnapper that since his life isn't in danger he isn't lowering the dragon teeth no matter what he does(clearly explain this on the bill board outside). It is highly unlikely that the kidnapper would now just start shooting or would attempt a kidnapping, knowing that none of the guards at the gate has the ability to open it. The gate guard opened the gate and the abandoned pickup was found 50km with a seriously injured guard besides it. He was shot in any case.

This also allows the remote security officer to activate a very large LED billboard, warning approaching vehicles to avoid the boom gate and the incident. As a general rule, guards must avoid close proximity contact with the public.

Estate circumvention
https://www.youtube.com/watch?v=pgJp8qnFPE4 in this video the driver's window was up so that the camera could not focus on his face. he opened the boom gates by sticking his hand out via the passenger window and typing in a password. Facial Recognition prevents unauthorized password usage. https://www.youtube.com/watch?v=Bxf4MeDLM74 use remote controlled LED spotlight and intercom to visually and audibly,depending on situation inform intruders that they have been spotted with Yolo. In almost 99% of cases they will make a retreat.

action analysis
Neural nets does action analysis, alerting that behavior isn't the norm. The videos are used for training the neural net.

https://www.youtube.com/watch?v=eppIM1LWppM aggressive behavior.

https://www.youtube.com/watch?v=5_l-fVxU8e4 running before robbing bank.

bkm watch
https://www.youtube.com/channel/UCcaO2ktnuTT8cH2pyVtrQvQ

i
A Yolo AI on Raspberry pi detects any type of weapon that the business owner is threatened with and automatically raises the alarm. Populate the area with hundreds of Raspberrys that will stream a live video over cell networks to indicate the escape route of assailants. The bandwidth requirements are very modest allowing non intrusive usage of nearby fiber via the inbuilt rasberry wify. The escape vehicle is unlikely to stop at red robots, allowing the Yolo system to flag it.

Any broke second year computer science student will be able to customize Yolo for your particular artificial intelligence project. The functionality has been so black boxed that with even rudimentary Linux and c skills a military grade project is construed.

A neural net can combine the audio to make an automated decision that a kidnapping or carjacking is taking place. Neural training is done on the Unreal gaming engine where real video is lacking. There might not be enough YouTube videos of people jumping over walls, but endless variations is possible with a gaming engine. Having virtual-world representations of real-world scenes makes it relatively easy to generate training data for deep learning networks because once the virtual scenes are created, they can be duplicated and manipulated in any number of ways.

If they kill the kidnap victim, well they would have killed him in anycase. Always fight back and if you die, you would have died any way. A Tracked vehicle robot drags a Net_gun spiked net across the road bringing the vehicle to a standstill and engaging it with a PID controld,   Gimbal,   PepperBall. A Yolo controlled ground robot or uav engages first with plain paintball, hopefully the intruder starts shooting with a firearm, allowing return fire with two centimetre rubber or nylon balls, just one shot off these will incapacitate. Under SA law you always have to use proportional force. Security companies protecting the copper loads on trucks from Zambia wear a bodycam to prove that when they shot with a pepper ball gun into the backs of assailants, it was because they turned their backs at that moment. the angled plates of the inspector tactical bot(picture right side) prevents bullet penetration, especially if made out of titanium (see Tracked vehicle).

In South Africa a mob of three hundred people marched to an area manager of the post office and assaulted him. Had he been aware that 300 people were walking down the street he could have locked his gates in time. Deploying Yolo to automatically count the average pedestrian traffic and alert when this count is exceeded is trivial.

incidents
https://www.youtube.com/watch?time_continue=5&v=5FGjWgHr0aQ gate forced open

https://www.youtube.com/watch?v=pI5hcsdnDMg taxi 2:10min the gear on the gate opener is broken. Taxi drives by, then reverses back. The yolo bounding box surface area will get smaller, stop then increase in size as the taxi reverses. It is a crude way of detecting anomalous behavior, the Nvidia Jetson developer activity analysis is more robust.

https://www.youtube.com/watch?v=To70_U2w_Q8 carjacking after driving into business. Dragon teeth activating before after the driver crosses perimeter would help. Installing a sentry tower armed with flame thrower and firing co2 Net gun Napalm cannon rounds is another solution, but obviously impractical.

https://www.youtube.com/watch?v=C6CiYVHtpPw Armed robbery of family arriving home. Neural net behavior analysis sends an automated alert. For example program Yolo to only send an alert when it detects people around a vehicle, without the doors opening first and driver climbing out and combine this with some Facial Recognition cloud based app like Faceter. Wouldn't a Tracked vehicle with turret, Yolo tracking pedestrians outside the residence be at least some sort of a deterrent? Obviously for safety reasons it would only be fitted with a Net gun. Yes, ofcourse the robot will probably get stolen, this is why you need to be able to build them yourself with a Lathe after watching the videos from Machine design.
 * https://www.youtube.com/watch?v=E_bQtXcNjp0 Car jacking attempt
 * https://www.youtube.com/watch?v=MZNgdODgu-0 another car jacking
 * https://www.youtube.com/watch?v=5svF6X2Gbdw car jacking.
 * https://www.youtube.com/watch?time_continue=14&v=BE12ED_nPrc armed robbery KZN, 13 old killed.
 * https://www.youtube.com/watch?v=yLnqfpmMjLA car pins down from behind. Drive into a cordon of Tracked vehicle robots that will engage with nylon and PepperBall.
 * https://www.youtube.com/watch?v=QT8PGCWG0mY Inside job robbery the gate was manually operated and left open: automate the gate.
 * https://www.youtube.com/watch?v=JefTJodtOsM kidnapping people and then forcing them to withdraw cash at 3min. 7:50min the classic maneuver of pinning an exiting vehicle down. Yolo can easily send an automated alert by noting people disengaging from the second vehicle. Stack a few streaming Raspberry pi's around the exit with Yolo scanning the road for approaching vehicles. If possible reverse back into the driveway so that existing is easier.
 * https://www.youtube.com/watch?v=8BkUfQB-rtA
 * https://www.youtube.com/watch?v=NCDTClmES30
 * https://www.youtube.com/watch?v=JU1kU1DpfFE Yolo draws slow moving bounding box around car, identifies make and model with a Titan X Nvidia on Faceter cloud. Initially a reduced dataset Yolo quickly identifies that some sort of vehicle is at the gate, a larger dataset in the cloud classifies the vehicle. Intruder steps out of car, combine the two events, a stationary bounding box around car and person climbing out(bounding box around him) for a crude behavior analysis event. The time it took them loitering and staking out was enough for a response. AutomaticGateOpeners z-Axis mechanism prevents forced opening of a gate. Gate openers are made to open gates, not to secure them. Around 1:30,the burglar removes the plastic gate opener covering, shorts something and effortlessly opens the gate!. The public must be willing to pay and demand from manufactures an avoidance mechanism. Because people are not really willing to pay for proper security, the manufacturers gives the client what he wants: a cheap easily compromised system. The Yolo cloud Titan X detects any attempt at cutting through the gate with angle grinder or cutting torch. A raspberry pi doesn't have the Cuda cores processing speed to detect all possible means of intrusion, it is used primarily for person and vehicle detection.

steel bar
https://www.youtube.com/watch?v=J8x2PnVeZaM at 2:30 burglars easily break open a gate opener with the leverage a steel rod provides. Have an unarmed(don't kit it out with a Flame thrower!) Tracked vehicle UGV steer automatically to the gate to address the intruders and audibly inform them that security is on its way, in most cases this will be enough to prevent a break in. The number plate and faces can be extracted with the latest Neural papers with code breakthroughs in AI.

https://www.youtube.com/watch?v=J8x2PnVeZa https://campbellwindowfilm.com/security-windows-doors/3m-window-film-commercial-building-security/#Riot%20Glass

https://www.news24.com/Video/SouthAfrica/News/watch-shots-fired-at-guard-during-bedfordview-hijacking-20171002 In this video a women is ejected and her car taken. The Yolo system does realtime face and object recognition counting how many people are in the car and who they are. If it detects a person other than the designated driver or additional occupants, indicating a kidnapping it cuts the engine at a busy robot intersection using stored GPS coordinates, reducing the chance that the owner is executed out of anger. Crime can easily be stopped, it needs an awareness of the free technology available. In high risk situations such as farming, the bodycam continually MeshNetworking streams the surroundings to a Jetson GPU. Only on Yolo detecting a person will it relay the situation to the Faceter behavior analysis cloud network, reducing considerably the bandwidth usage and GPU computing costs.

https://www.youtube.com/watch?v=PnrXlcrE-_I from https://www.cctvlink.co.za/browse-our-shop/1-3-mp-intelligent-people-counting-wdr-network-camera/ product uses Yolo sells for ZAR13000. Raspberry pi and Gprs and wifi streaming is a bit cheaper.