Loading Project Media
Live
CCTV Inference
<1s
Alert Latency
CV
Vapour tracking
(Overview)
ThisprojectisanadvancedcomputervisionsystemdeployedonliveCCTVstreamstomonitoranddetectvapinginrestrictedareas.Bycombiningpostureestimation(hand-to-mouthtracking)withvapourclouddetection,theMLpipelineaccuratelyflagspotentialincidentsandstreamsdetailedalertsdirectlytoWhatsAppwithconfidencescoresandsignalbreakdowns.
(The Challenge)
DetectingvapingfromadistanceonstandardCCTVcamerasisnotoriouslydifficultduetolowresolution,smalldevices,andvaryinglightingconditions.Thesystemneededtominimizefalsepositiveswhileaccuratelyinterpretingcomplexsignalslikewrist-to-mouthdistanceandinhalationpatternsinreal-time.
(The Solution)
Implementedacustom-trainedobjectdetectionandtrackingmodelusingPyTorchandOpenCV.Theinferenceengineprocesseslivevideofeeds,applyingheuristicslike"hand_near_mouth"and"vapourconfidence".Whenthealertthresholdisbreached,anautomatedwebhooktriggersadetailedWhatsApppayloadforadministratorstoinvestigate.
- 01
DevelopedthecomputervisionpipelineusingOpenCVandPyTorchtoprocessliveCCTVfeeds.
- 02
TrainedcustomMLmodelstodetectspecificbehaviouralsignalslikehand-to-mouthproximityandvapourclouds.
- 03
Engineeredthereal-timealertingsystemintegratingwiththeWhatsAppAPIforinstantadministratornotifications.
- 04
Optimizedtheinferenceenginetorunefficientlyonedgedevices,maintaininghighframerates.
Built with
PythonPyTorchOpenCVYOLOWhatsApp API
Haveaprojectinmind?
kryptochaingames@gmail.com