Computer Vision / Edge AI Engineer: Real-Time Shoplifting Detection for Pharmacies (Veesion.com-like system)

@murtaza1122

Hi, I have recently worked on CCTV for in house users activity tracking like how much they productive are, I have also worked on overhead bridges CCTV for transport safety. I can provide a POC before we start the project and if it seems ok we can proceed. Also if you can share that if multiple cameras are overlapping a zone or only one camera for one zone? I can communicate in English but never tried French, but I can use translators where required. Lets discuss

@sunainawbst

Computer Vision | Edge AI | Motion Tracking | Mobile Apps Hi, Thanks for sharing the details. I have experience in AI applications, motion tracking, mobile apps, and backend development. I can help build your CCTV-based detection system, from camera integration and AI processing to real-time alerts and mobile apps. My approach would be to start with the paid test, validate live camera streaming and person tracking, and then expand to the full MVP with gesture detection, alert verification, clip generation, and a feedback system. False alarm reduction: I would combine object tracking, pose estimation, and action sequences to distinguish normal phone use from possible concealment. Alerts would be triggered only after checking the full sequence, not a single movement. 30-camera setup: I would propose a GPU-powered edge server with video ingestion, AI inference, tracking, and alert services. The final hardware would be selected after testing actual camera streams and performance. I don’t have a verified production deployment handling 30 CCTV cameras, so I would validate this during the paid test before committing to the full setup. Thanks! Purnima

@sajjadtaghvaeifr

Hello, I trust you're doing well. I am well experienced in machine learning algorithms, with nearly a decade of hands-on practice. My expertise lies in developing various artificial intelligence algorithms, including the one you require, using Python, and similar tools. I have worked with pytorch, and tensorflow to develop DL models, .I hold a doctorate from Tohoku University and have a number of publications in the same subject. My portfolio, which showcases my past work, is available for your review. Your project piqued my interest, and I would be delighted to be part of it. Let's connect to discuss in detail. Warm regards. please check my portfolio link: https://www.freelancer.com/u/sajjadtaghvaeifr

@sufyanasmat38

Hello, I have strong experience with computer vision, live video processing, object detection, tracking, pose analysis, edge AI, and mobile alerts. I can build the pipeline from RTSP streams through detection and behavior verification to real time mobile notifications, with a strong focus on reducing false positives and privacy by design. I can also handle the edge architecture, API, feedback driven retraining, and iOS Android alert workflow. I can start with the paid Phase 1 test and provide the requested technical architecture and relevant examples.

@a4logic

Hello, With my 15+ years of experience developing high-quality and scalable web, mobile, AI, and automation solutions, I am confident I will be a perfect fit for your real-time shoplifting detection project at Veesion.com. I am skilled in edge AI deployment at scale and have a proven track record of processing live CCTV camera streams efficiently. My specialties in AI integrations, third-party APIs, and cloud deployment make me proficient in reducing false positives - an essential aspect of any computer vision project. I understand the criticality of privacy-by-design in your project; hence my approach aligns with your vision to comply with Quebec Law 25 / GDPR principles. My extensive knowledge of both server-side and frontend technologies will assist in creating a robust, secure environment for your pharmacy-client app. Over-and-above, my good command over written English and French (as a plus) facilitates seamless communication with the team. My number one goal is client satisfaction, which has led me to build strong relationships with my clients in the past. Perhaps most importantly, my strong work ethics and demonstrated abilities can ensure full ownership of the work-for-hire nature of this project. I'm excited to embark on this challenging journey with you whereby we can revolutionize real-time shoplifting detection for pharmacies together! Thanks!

@emanz03

I can build the Phase-1 pipeline around RTSP/ONVIF ingestion, real-time person detection/tracking, event-triggered clip buffering and mobile push alerts, then evolve it into the full edge architecture. For live video work, I have worked with real-time CV/video pipelines and edge-oriented inference; I can demonstrate relevant implementation privately rather than overstate a 30-camera pharmacy deployment I have not yet delivered. For false alarms such as a customer pocketing their own phone, I would combine object/hand tracking, pose trajectories, shelf interaction, item provenance, temporal context and a second-stage verifier instead of triggering on a single hand-to-pocket motion. For 30 cameras, I would use an x86 edge server with an NVIDIA GPU, GStreamer/DeepStream-style ingestion, batched inference, detector + tracker + pose/temporal models, local clip buffering, and an API/mobile alert layer. I’m comfortable starting with the paid test and proving latency, stability and alert quality first.

@invoketech

HELLO, WE HAVE WORKED ON SIMILAR REAL-TIME CCTV, COMPUTER VISION AND EDGE AI SYSTEMS AND CAN PROVIDE RELEVANT EXAMPLES. I have carefully reviewed your requirements. We have 10+ years of experience with Python, OpenCV, YOLO, object tracking, pose/action recognition, RTSP/ONVIF, edge AI and mobile apps. We can build the complete pipeline from camera streams to verified alerts, video clips and mobile notifications. Working Flow: Camera Integration → RTSP/ONVIF/NVR streams. AI Detection → People, objects, tracking and gestures. Verification → Temporal analysis and rules to reduce false alarms. Edge Deployment → Optimized inference for up to 30 cameras. Mobile & Backend → Alerts, clips, feedback and monitoring. Retraining → Staff feedback to improve future models. For false alarms, we would combine object tracking, hand/object movement, pose, zones and multi-frame behavior rather than triggering on a single frame. We can start with the paid Phase 1 test and scale to the MVP. Thanks, Invoke Tech

@msaqibimran

Hi, I can build a real-time shoplifting detection system that ingests live streams from existing CCTV setups and tracks movements effectively. Based on my understanding, the initial phase could be executed in about 2 weeks, focusing on people detection and tracking. To clarify, do you have specific camera models in mind, and is there a preferred edge device for processing? Additionally, how do you envision the alerting mechanism for staff before alerts are sent out? For the systems architecture, I’d utilize a powerful edge device like the NVIDIA Jetson platform, optimized for handling multiple streams, ensuring low latency and high efficiency. This keeps the processing local while reducing bandwidth usage. I've worked on similar implementations, leveraging object detection and tracking frameworks to achieve robust performance in live environments. Happy to discuss the current setup and take it from there. Best regards, Mohammad

@sanjana22kr

Hello there , I can build your complete real-time CCTV → Edge AI → Behavior Detection → Alert → Mobile pipeline. 1. Live CCTV: Experience with RTSP/IP streams, GStreamer/FFmpeg, object detection, tracking and real-time alerts. 2. False positives: Use person/object tracking + pose/keypoints + temporal behavior analysis + zones/rules, rather than triggering on a single frame. Staff feedback can improve the models through retraining. 3. 30 cameras: I’d use an NVIDIA GPU edge appliance with TensorRT/CUDA, hardware decoding and optimized multi-stream inference. Architecture: 30 RTSP → Detection → Tracking → Action Recognition → Verification → Clip → API → Push. Stack: Python/C++, GStreamer, YOLO, ByteTrack, TensorRT, Node/NestJS or FastAPI, PostgreSQL, Flutter. I can start with your 1–2 week paid Phase 1, then scale to the full MVP with complete source-code and documentation handover. Thanks

@tobiask5

Hello, This is an edge video-intelligence pipeline, not just an object-detection model. The core challenge is combining RTSP ingestion, tracking, pose/action recognition, temporal verification, and low-latency alerting reliably across up to 30 cameras. I’d first build the paid test around one RTSP stream: person tracking → defined event detection → clip generation → push alert. For the MVP, I’d then move inference to edge hardware, add zone/rules logic and false-positive verification, and connect the mobile app, backend, camera monitoring, and feedback-driven retraining loop. The main risk is false positives in normal customer behavior, so alerts should require temporal/contextual evidence rather than a single frame or gesture. For 30 cameras, are you already committed to specific edge hardware/NVR models, or should the hardware architecture be proposed as part of Phase 1? Thanks, I look forward to hearing from you. Tobias

@jagratip3

Hi there, we can build the complete edge-AI CCTV pipeline, from RTSP/ONVIF ingestion and detection/tracking to behavior analysis, alert verification, mobile notifications and feedback-based retraining. We have experience with Python, OpenCV, YOLO, pose estimation, tracking, APIs, cloud and edge deployment. We can start with your paid Phase 1 test and offer a free proof-of-concept demo. Happy to answer all 3 technical questions and share relevant work. Thanks!!

@CorporateRD

Good day, thanks for sharing your project details, Here are answers to your questions: 1. False-alarm reduction: I’d combine person/object tracking, hand-pose trajectories, temporal action recognition, scene/zone context, and a verification layer so normal actions like putting a personal phone in a pocket are filtered before an alert is triggered. 2. 30-camera architecture: I’d use a GPU-accelerated edge server such as an NVIDIA-based industrial PC with TensorRT, using a shared RTSP/ONVIF ingestion pipeline, batched inference, multi-object tracking, and lightweight pose/action models to process up to 30 streams in real time while keeping video local. What I understand that you need to build an end-to-end real-time AI shoplifting detection platform for pharmacies, connecting existing CCTV/NVR streams to edge AI for person detection, tracking, pose and gesture analysis, followed by verified alerts with short video clips in a mobile app. My questions: - How should the system distinguish a customer putting their own phone or personal item into a pocket/bag from a genuine concealment event, and what contextual signals should be used before generating an alert? - Should each store be able to configure camera zones, gesture sensitivity, severity levels, and enabled/disabled detection types independently? Regards, Royal Designs Budget and Duration are TBD>

@emilesd617

1. Live camera streams, you can check camera realted task in my profile. 2. Phone in pocket: I'd track where each object comes from for every person ID. An alert fires only when a product class was picked up in a shelf zone and then disappears near a pocket, bag or clothing. A phone that was already in hand, or never came from a shelf, is ignored. On top of that, a second-stage clip classifier checks the full gesture sequence before anything reaches staff, and "Not theft" feedback goes back into retraining. 3. 30 cameras: an NVIDIA Jetson AGX Orin (or an x86 box with an RTX-class GPU if more headroom is needed) running DeepStream. Video decoding is the main bottleneck at this scale, and the hardware decoder handles it. I'd use camera substreams at 10–15 fps with a batched TensorRT detector and tracker for all cameras. Pose and action models would run only on people inside active zones, which keeps GPU load low. Everything would run in Docker with remote updates and camera health monitoring. I'm an individual freelancer and can work on any time zone you want. Please contact me with the best time for you to have a quick chat. Looking forward to discussing more details. Thanks. Emile.

@modesorehari6

⭐⭐⭐Hi there, From Your Vision To Successful — I will take responsibility⭐⭐⭐ I can build a reliable real time computer vision pipeline for CCTV based shoplifting detection. I have relevant experience with live video processing, object detection, multi object tracking, pose analysis, edge AI and mobile applications using technologies such as Python, OpenCV, YOLO, tracking pipelines, APIs and React Native. For Phase 1 I would connect RTSP or ONVIF streams, detect and track people with persistent IDs then trigger event based clip creation and push notifications. For the MVP I would design an edge first architecture with optimized inference, zone and rules configuration, false positive verification, camera health monitoring, backend APIs and iOS and Android alerts. For false alarms such as a customer putting their own phone in a pocket I would combine object tracking, hand and pose movement, temporal action analysis and context based rules instead of triggering from a single frame. I can start immediately as I am free now. I will follow your technical suggestions and milestones closely and can work with your team long term as the system moves from the paid test to production. Thanks

@FreeLancers360

As a seasoned Full Stack Developer, specializing in AI integration and Mobile App Development, I strongly believe I can craft a state-of-the-art solution for your shoplifting detection system. Throughout my career, I have built intricate, scalable software products tailored to numerous business requirements. My extensive understanding of Computer Vision, edge-infrastructure and knack for mitigating false alarms makes me the perfect fit for this complex project. To ensure I am well-prepared to tackle this demanding task, I put in place an AI-powered shoplifting deterrent solution fueled by live camera streams from multiple pharmacies. This experience included integrating countless CCTV feeds onto one central device while maintaining real-time performance tailored specifically for each store's needs. The solution was grounded on reducing false positives by thorough filtering algorithms that understand gesture context. This approach aligns perfectly with your privacy-by-design specifications. For your current scope, the proposed hardware configuration comprises an optimized setup with multi-threading architecture and robust edge deployment capabilities in-built to handle in-store inference on up to 30 cameras simultaneously. This design offers the advantage of significant cost-saving while

@malikabdulbasit3

As an experienced and versatile computer vision and mobile app developer, I bring a unique skillset to the table in tackling the scope of your project. Over the years, I’ve successfully developed applications similar to what you're looking for including e-commerce apps, real-time apps, and even business dashboards. In terms of delivering real-time image analysis and app solutions optimized for performance, my work in developing live tracking systems and implementing Firebase's notification service puts me in a strong position to handle your project requirements end-to-end. Additionally, my proficiency in REST API integration and backend connectivity is key to realize an efficient solution for your video ingestion, multi-object tracking, and alert generation. Privacy-by-design is not just a buzzword for me. I'm adept at implementing it within applications keeping in mind local data protection legislation such as Quebec Law 25 / GDPR principles. Moreover, French-English bilingualism allows me to understand conversations like this pitch fluently easing communication barriers if any. To reduce false alarms when customers put their personal items in their pockets or bags, I propose leveraging my knowledge of pose estimation - specifically interpreting hand movements. Let’s collaborate on this transformative endeavour to make shopping spaces safer from malicious intents! Chat with me để trò chuyện!

@smitk9999

I understand you want a system that watches your pharmacy's existing cameras, spots someone hiding a product in their clothes, bag or backpack, and sends staff a short clip on their phone within seconds, running on one in-store box for up to 30 cameras with no facial recognition, starting with a paid test on a single live stream. I've built Python computer vision pipelines on RTSP streams with YOLO detection, tracking and pose estimation, and to stop phone-in-pocket false alarms I'd only flag a pocket gesture when that same hand was tracked picking something off a shelf first. I can start work right away and contact me to discuss more details. Thanks. Smit

@GridsmithLTD

Hello, I reviewed the architecture, and the core challenge is achieving reliable real time behavior recognition across many CCTV streams while keeping edge inference efficient and false alerts low enough for staff to trust the system. For live camera processing, I have experience with computer vision pipelines involving RTSP streams, object detection, tracking, and event based video processing, and can share relevant project evidence in chat. For false alarms such as a customer putting their own phone in a pocket, I would combine object tracking, pose and hand trajectory analysis, temporal action recognition, zone context, and a verification layer rather than triggering from a single frame. For 30 cameras, I would evaluate an NVIDIA edge GPU based architecture with optimized TensorRT inference, centralized tracking and event processing, local clip buffering, and a resilient service layer for remote monitoring and updates. Also a quick question, are the existing NVRs providing direct RTSP streams with hardware accelerated decoding, or would the edge module need to handle the full camera decoding workload? Best Regards, GridSmithLTD

@manh2906

Hi There - Manh here "REAL-TIME SHOPLIFTING DETECTION" — you need reliable CCTV behavior detection that can run on edge hardware with low false alerts. I would separate the pipeline into stream ingestion, person tracking, pose/action analysis and a verification layer, so one frame or gesture does not immediately trigger an alert. For 30 cameras, I’d benchmark inference speed, memory and stream load on the target hardware before fixing the deployment architecture. For the phone-in-pocket case, I’d use temporal behavior signals and context such as zones and object interaction rather than treating the hand movement alone as theft. Can you provide a sample RTSP stream and confirm the edge hardware available for the Phase 1 test? Looking forward to working with you.