Fare­well, Fal­se Alarms

pro­mi­seQ is re-thin­king the secu­ri­ty indus­try with AI-sup­port­ed video surveillance

A shadow, a moving ani­mal or rain - anyo­ne who reli­es on video sur­veil­lan­ce to secu­re their buil­dings or com­pa­nies is often faced with fal­se alarms. The flood of video data gene­ra­ted by the­se sys­tems poses a major chall­enge for secu­ri­ty staff. Alarms are sent unfil­te­red to the emer­gen­cy con­trol cen­tres whe­re they then have to be veri­fied by humans. The rate is high: over 98% of the alarms trig­ge­red are harm­less. But they cost time, money and resour­ces. In the worst-case sce­na­rio, they not only lead to unneces­sa­ry deploy­ments - they can also dis­tract from real thre­ats. If employees have alre­a­dy cli­cked a cat away on 99 occa­si­ons, they will do the same the hundredth time - except that now it’s a real emergency. 

The Media­Tech Hub start­up pro­mi­seQ has set its­elf the goal of sol­ving this pro­blem with the help of arti­fi­ci­al intel­li­gence (AI). promiseQ’s AI-based plat­form ana­ly­ses video data in real time and auto­ma­ti­cal­ly fil­ters out fal­se alarms. This allows secu­ri­ty staff to focus on the essen­ti­als and be able to reco­g­ni­se and respond to genui­ne thre­ats more swiftly. 

promiseQ’s AI plat­form is based on a deep lear­ning algo­rithm that was trai­ned with mil­li­ons of video data. The algo­rithm is able to reco­g­ni­se dif­fe­rent pat­terns and objects in vide­os and thus distin­gu­ish bet­ween genui­ne and fal­se alarms. Genui­ne alarms are for­ward­ed, while the others are fil­te­red out. But that’s not all. As foun­der Tol­ga Ermis explains: “We have deve­lo­ped a holi­stic plat­form around the AI fil­ter, a video intel­li­gence plat­form. You can use it to con­nect to came­ras live, mana­ge cus­to­mers and pro­per­ties, and iden­ti­fy and chan­ge blind spots or out-of-focus set­tings. If someone has chan­ged the direc­tion of the came­ra or cover­ed it with paint or glue, this can also be quick­ly iden­ti­fied using stored refe­rence images.” 

Came­ra sys­tems are trai­ned using exter­nal AI hardware

Tog­e­ther with his part­ner Eli­as Kar­del, Ermis had alre­a­dy work­ed on auto­ma­ted dri­ving pro­jects for car manu­fac­tu­r­ers, focu­sing on such sub­jects as crowd­sour­cing for AI trai­ning and came­ra-based lane kee­ping for moving cars. They came tog­e­ther to launch pro­mi­seQ with the idea of com­bi­ning AI and crowd­sour­cing - i.e. with the par­ti­ci­pa­ti­on of humans. The team now com­pri­ses over 30 employees who are con­stant­ly deve­lo­ping the idea. The focus is no lon­ger on the human review fac­tor, i.e. the cross-check bet­ween AI and crowd­sour­cing, but this is still being offe­red as a pre­mi­um model for the AI fal­se alarm filter. 

They have just laun­ched a new pro­duct, the pro­mi­se­Qu­be. This brings real-time hard­ware in a box and enables exten­si­ve AI fea­tures to be added to secu­ri­ty sys­tems on-site. “We can make any ’stu­pid came­ra’ ultrasmart with the box,” Tol­ga Ermis says. The Qube reco­g­ni­s­es the various object clas­ses - the­se are curr­ent­ly, for exam­p­le, peo­p­le and moving vehic­les. Inter­nal tests are being under­ta­ken with the zero-shot model used in machi­ne lear­ning to see how a varie­ty of object clas­ses could be defi­ned in future. Depen­ding on such spe­ci­fi­ca­ti­ons as “white trai­ners”, “peo­p­le with masks”, “peo­p­le with blue pull­overs”, “peo­p­le with hel­mets” - or very cri­ti­cal objects like fire or peo­p­le car­ry­ing knives. 

The high-per­for­mance hard­ware gives secu­ri­ty com­pa­nies unpre­ce­den­ted fle­xi­bi­li­ty. They had pre­vious­ly had to rely on came­ra manu­fac­tu­r­ers to spe­ci­fy the alarm defi­ni­ti­on. If you wan­ted to fil­ter out object clas­ses such as fire or vehic­les with cer­tain licence pla­tes, you were having to rely on what the manu­fac­tu­r­ers offe­red and their trai­ning of the came­ras via tra­di­tio­nal neu­ral net­works. As Ermis explains: “Such tra­di­tio­nal neu­ral net­works are like litt­le child­ren who have to be shown ever­y­thing a thousand times. For exam­p­le: here’s a pic­tu­re of a kni­fe, here’s one of a kni­fe in the dark, here’s one of a fork, here’s a kni­fe from the side ang­le. It takes huge amounts of data for the came­ra to be able to react to the cor­re­spon­ding situation.” 

A game chan­ger, not only for the secu­ri­ty industry

The zero-shot model is curr­ent­ly being updated for the pro­mi­se­Qu­be - pre-trai­ned with the free­ly available know­ledge of the enti­re Inter­net. It can now be com­pared to a uni­ver­si­ty gra­dua­te ins­tead of to a small child, Ermis sug­gests. The Qube can pro­vi­de came­ras with con­text and knows what the objects it is sear­ching for look like in hundreds of varia­ti­ons. The spe­ci­fi­ca­ti­ons are made in natu­ral lan­guage: “Plea­se reco­g­ni­se peo­p­le in blue pullovers.” 

The pro­mi­seQ team is also working on the pos­si­bi­li­ty in future of using this to con­trol and acti­va­te doors, loud­spea­k­ers, locks or fog machi­nes. “You could descri­be the pro­mi­se­Qu­be as being like Amazon’s Ale­xa, but with tooth and nail,” Ermis says. 

In future, pro­mi­seQ would like other sec­tors to have access to such fle­xi­ble came­ra spe­ci­fi­ca­ti­ons which are pos­si­ble thanks to zero-shot machi­ne lear­ning. The retail sec­tor would be a logi­cal part­ner. Which ais­les are the busie­st? What are the cus­to­mers’ demo­gra­phics? AI-sup­port­ed video sur­veil­lan­ce could pro­vi­de a swift ana­ly­sis and then auto­ma­ti­cal­ly check stock levels. At the moment, though, the start-up is focu­sing on revo­lu­tio­ni­s­ing the secu­ri­ty indus­try. Demand for the first pro­mi­se­Qubes is so high that the­re is now a wai­ting list for the next pro­duc­tion run which is plan­ned for ear­ly summer. 

About MTH Blog

The media technologies of the future are already being used today – not only in the entertainment sector, but also in a wide variety of industries. Christine Lentz meets up with tech enthusiasts, established companies and researchers for our monthly MediaTech Hub Potsdam blog to tell the stories behind the innovative business models.