Pretrained Video Analytics for People, Vehicles, Retail and Safety
Ready-trained detectors and classifiers for cameras that count people, read plates, check safety gear and watch shelves.
Tier Grades
Tier A: Frontier AI labs
stretch
Chip vendors such as Intel, NVIDIA and Qualcomm publish these models to sell hardware, not frontier labs.
Tier B: Indonesian AI labs
feasible
Small models of about 1.8M parameters are easy to retrain on Indonesian footage.
Tier C: Institution units such as ITB
feasible
A university unit can retrain them, and the 2-author CPPE-5 paper shows dataset work at that scale.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
Ready-trained camera models give fast proofs of value in traffic, retail and safety.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeIntel person-detection-0200: 1.817M parameters and 0.786 GFLOPs at 256x256 input; training compute not given.
- DataInternal test set (AP 0.2398 at IoU 0.50:0.95); training data size not given.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 2 authors on the CPPE-5 dataset paper (arXiv 2112.09569).
- Time horizonMonths.
- Shelf-life signalsVendor model zoos are replaced by open-vocabulary detectors, Privacy rules limit people analytics
- Pipeline roleVendors release model zoos; Tier C retrains them on local footage; Tier D deploys them on client cameras.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
PeopleNetvisionexemplar
Intel person detection modelsvisionexemplarLPDNet and LPRNetvisionexemplar
PPE-Detection (GearGuardNet)visionexemplarTrafficCamNetvisionproduced byInspireFacevisionproduced byIntel face modelsvisionproduced byIntel person attributes and re-identificationvisionproduced byIntel licence plate recognitionvisionproduced byIntel instance and semantic segmentationvisionproduced byIntel product detectionvisionproduced byNVIDIA TAO pretrained detection and segmentation modelsvisionproduced byDashCamNetvisionproduced byVehicleMakeNet and VehicleTypeNetvisionproduced byFaceDetect and FaceDetectIRvisionproduced byNVIDIA TAO human understanding modelsvisionproduced byRetail Object Detection and Recognitionvisionproduced byPerson-Foot-Detectionvisionproduced byLinks
R&D TopicPretrained Video Analytics for People, Vehicles, Retail and Safety
ModelsPeopleNetNVIDIA, visionIntel person detection modelsIntel, visionLPDNet and LPRNetNVIDIA, visionPPE-Detection (GearGuardNet)Qualcomm, vision+14 more
PeopleNetNVIDIA, visionIntel person detection modelsIntel, visionLPDNet and LPRNetNVIDIA, visionPPE-Detection (GearGuardNet)Qualcomm, vision+14 more
Models 18
DashCamNetNVIDIA, visionproduced byFaceDetect and FaceDetectIRNVIDIA, visionproduced byIntel face modelsIntel, visionproduced byIntel instance and semantic segmentationIntel, visionproduced byIntel person attributes and re-identificationIntel, visionproduced byIntel product detectionIntel, visionproduced byNVIDIA TAO human understanding modelsNVIDIA, visionproduced byNVIDIA TAO pretrained detection and segmentation modelsNVIDIA, visionproduced byPPE-Detection (GearGuardNet)Qualcomm, visionexemplarRetail Object Detection and RecognitionNVIDIA, visionproduced byInspireFacevisionproduced byIntel licence plate recognitionIntel, visionproduced byPapers
- NoScope: Optimizing Neural Network Queries over Video at ScaleDaniel Kang et al., arXiv preprint, 2017 founding
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision ApplicationsAndrew G. Howard et al., arXiv preprint, 2017 founding
- Edge Intelligence: Paving the Last Mile of Artificial Intelligence with Edge ComputingZhi Zhou et al., arXiv preprint, 2019 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- raw.githubusercontent.com/openvinotoolkit/open_model_zoo/master/models/... https://raw.githubusercontent.com/openvinotoolkit/open_model_zoo/master/models/intel/person-detection-0200/README.md
- arxiv.org/abs/2112.09569 https://arxiv.org/abs/2112.09569
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Pretrained Video Analytics for People, Vehicles, Retail and Safety", https://aurion.semai.id/topics/edge-video-analytics/, records of October 2026.