Cameras vs. Sensors: How Not to Overpay for “Smartness” in Everyday Tasks

Almost all modern automation falls into two camps. In the first live honest, boring infrared sensors that stare at one tiny slice of reality and confidently answer a single question: “yes” or “no.” In the second camp are cameras with computer vision, which immediately demand “context, analytics, object behavior, and a heat map of humanity.” The baggage carousel at an airport is a perfect illustration of the difference. An IR sensor hangs over the spot where suitcases drop and, like a strict dispatcher, decides whether another bag can be released onto the circular belt. If another piece of luggage is passing through the landing zone, the sensor says “occupied” and slows the upper belt until a gap appears, meaning a free space. The result looks like logistics Instagram: suitcases lined up neatly, almost artistically, with no piles or pyramids of someone else’s underwear. But real life, as usual, is more complicated than the algorithm. Ten minutes later the belt is full of bags belonging to passengers stuck at passport control, and the system dutifully keeps spinning an endless loop of чужой luggage. For the IR sensor everything is perfect: no space, no accidents, the belt is moving. For passengers it’s a looping screensaver from hell, made entirely of other people’s suitcases. Only when humans step in, pull half the luggage off the belt, and suddenly create new gaps does computer vision get involved. The camera sees free zones, the algorithm understands that delivery can continue, and the system comes back to life. It’s a good symbol of the future of automation: simple sensors enforce local discipline, while computer vision tries to make sense of the bigger picture when reality crawls outside the original technical spec.

An infrared sensor in everyday use is like a guard sitting by a door, watching only the crossing line and honestly logging “someone passed, someone left.” It doesn’t care whether it was a person, a suitcase, a box, or a very motivated cat. It doesn’t build profiles, count people, or write logs. Its superpower is elsewhere: it almost never breaks, costs about as much as a good bag of coffee, and doesn’t need firmware updates. In an apartment building, stairwell, office corridor, storage room, or restroom, an IR sensor solves tasks at the level of “someone is in the zone, turn the light on; nobody’s there, turn it off.” It doesn’t need GPUs, neural networks, or DevOps. You install it once, set a timer, and forget about it. In a world where everything is being turned into a “smart subscription service,” this is a slightly old-fashioned but very sober model. IR sensors have limitations, and they don’t even try to hide them. They can’t tell people from animals, don’t know whether it’s an employee, a guest, or an intruder. They can react to drafts, a hot radiator, or a random heat spike. But in simple tasks they win honestly: when all you need to know is “is there an object in the zone or not,” deploying an entire army of computer vision often looks like trying to kill a fly with an orbital laser. IR sensors shine where the world still resembles a neat diagram on a blueprint. In corridors people mostly walk, in storage rooms someone occasionally enters, in a parking space there is either a car or there isn’t. In such scenarios the sensor doesn’t pretend to be smart, but it does its job so reliably that no one comes running at night asking, “why did it update and stop working?”

A camera with computer vision is no longer a “switch with a brain,” but a curious observer trying to understand what’s actually happening in the frame. It’s not enough to know that something moved; it wants to know who moved and why that matters. In an apartment, such a camera can tell a cat from a human, recognize that the homeowner came back rather than a stranger, and trigger the right scenario. In a parking lot it distinguishes specific license plates, not just the fact that a spot is occupied. In a store it counts people, separates staff from customers, and can even show where queues really are and where it only feels crowded. In the baggage carousel case, computer vision doesn’t see abstract suitcases but the structure of the flow: where jams form, where the belt is clogged with unattended bags, where passengers intervened and cleared space themselves. Where an IR sensor honestly says “no space, everything is occupied, deal with it,” a camera can understand that the situation is abnormal, the flow has stopped, and the belt has turned into an infinite replay of the same thing. Computer vision wins in a chaotic world, where there are many objects, they are different, behave unpredictably, and where the mere fact of their presence says very little. Of course, it has downsides: power, networking, storage, compute, updates, data protection, and ritual dances around privacy. But where a business needs to know “who, what, and how,” rather than just “there is movement,” a camera becomes not just a sensor but a data source for dashboards, meetings, and justifications for new budgets. Unlike an IR sensor, which at best blinks an LED and gets tired.

A home security system is the perfect battleground for simple sensors versus computer vision. The old approach is to put volumetric IR sensors in room corners, magnetic contacts on doors and windows, connect everything to a box that honestly screams with a siren if something moves at the wrong time. Simple, like a wired phone. Then the cat enters the frame. For an IR sensor, the cat is the same “living object” as a human, just smaller. The result is false alarms, disabled zones, and an owner who no longer knows whom they fear more: a burglar or the fluffy saboteur. A camera with computer vision looks a bit smarter here. It can learn to distinguish animals from people, understand that movements near the floor are probably the cat, while movements near the door handle are more interesting. It can recognize faces, know “friends” from “strangers,” and send you not just a “motion detected” push notification but a frame with a specific person. At the same time, the IR sensor remains useful as a trigger that wakes the system and says, “look here, something’s moving.” The camera takes on the role of investigator, figuring out who passed by, what they did, how long they stayed, and whether this is a case for sounding the siren or just logging the event. Home security stops being a binary “screams or doesn’t scream” system and becomes a more nuanced story with context. For example, a child got up at night to get water, not an unknown person creeping through the apartment. The IR sensor is the pocket alarm clock; the CV camera is the surveillance system that actually knows why it’s watching.

There are areas where computer vision looks great in presentations but loses in real life to a good old sensor that doesn’t film anyone or analyze anything. Parking lots are a classic example. In one scenario, a compact sensor hangs over each spot, understands “car or no car,” lights a green or red indicator, and a sign at the entrance shows “free spaces: 27.” The system knows nothing about plates or driver behavior, but it runs for years and rarely needs help. In another scenario, a camera under the ceiling sees dozens of spots at once, recognizes cars and plates, builds analytics, and can even suggest where to park. Beautiful and convenient, but more expensive and complex. In restrooms and utility rooms the story is even harsher. The question isn’t analytics but basic automation: turn the light and ventilation on when someone enters, and turn them off after a while. An IR sensor solves this perfectly without capturing images or raising endless privacy questions. Trying to do the same with video surveillance is like taking out a loan for a supercar just to drive to the corner store for bread. In stores, IR barriers at entrances often count how many people entered and exited. Computer vision can do more: distinguish staff, couriers, and customers, build heat maps, estimate queue lengths. But if you just need a rough “how many people came in today,” an IR barrier will handle it without servers or luxurious infrastructure. There are places where cameras truly shine, and others where simple sensors have long claimed their niche and feel perfectly at home.

On the factory floor, the romance of AI quickly disappears and turns into harsh engineering discipline, where mistakes are measured in injuries and fines, not likes. On conveyors, IR sensors control not only presence but process correctness. On sorting lines they say “there’s a box here, you can push it to the side chute” or “the belt space is free, you can drop the next item.” They don’t know what’s inside the box, and they don’t need to. Their job is simple: don’t let the mechanics work blind. When human safety is involved, IR or laser curtains come into play. This is no longer about convenience but about life and health. If a person’s hand enters a dangerous zone, the beam is broken and the machine stops instantly. Such a system can be certified, tested, documented, and signed off by an engineer responsible for safety. Try doing the same with a camera and a neural network that can occasionally go blind because of dirt, glare, or a poorly timed firmware update. Computer vision on the factory floor makes sense where you need to see more than “hand in, hand out”: quality control, defect detection, counting boxes, analyzing line utilization, recognizing object types. But at critical points where a mechanism must stop reliably when a boundary is violated, simple sensors are still more reliable and, importantly, more understandable to inspectors. As a result, a modern factory often looks like a hybrid: IR curtains and sensors handle safety, while cameras and CV help management understand that production is not only fast but efficient.

On paper, computer vision often looks like a universal answer to everything. One camera sees it all, a neural network magically extracts events, and you enjoy beautiful reports. In the real world, you pay not just for hardware. A camera needs a network, stable power, video storage, an analytics server, updates, and a person to manage it all. You have to think about what happens when the connection drops, what about privacy, how long to store archives, who has access, and what happens when someone replaces the lighting fixtures and all your tuned analytics start seeing ghosts. An IR sensor is painfully boring but honest. It has no firmware to accidentally update on a Friday night. If it breaks, it usually just stops responding, which is relatively easy to diagnose. It stores no data, can’t be hacked for blackmail, and raises fewer eyebrows from auditors and lawyers. That doesn’t mean cameras and computer vision are always “expensive and complex.” One camera with good analytics can easily replace dozens of point sensors, especially where you need to count people, monitor zone occupancy, and recognize object types at the same time. The key is not to fall in love with the idea that “one camera will solve everything,” and to remember that reality loves dirt, glare, clogged networks, and people who casually turn off power “to that weird little box, it’s probably not important.”

If you don’t want to hold a strategic session with engineers, lawyers, and neural network enthusiasts every time, remember one simple rule. If the question is “yes or no,” “occupied or free,” “someone is in the zone or no one is,” it’s almost always cheaper and more reliable to use an infrared sensor. It’s perfect for corridor lighting, parking space occupancy, starting a conveyor when an object appears, or stopping a machine when a boundary is crossed. Where the question sounds like “who exactly, doing what, how many of them, how long has this been happening, and is this normal at all,” simple sensors hit their ceiling fast. If you need to understand who’s standing in line and how many cashiers are serving them, why a baggage carousel turned into an infinite loop of чужой suitcases, who walked around your apartment while you were away, or how store space is actually used, cameras and computer vision become unavoidable. The best systems, ironically, don’t choose between sensors and cameras; they combine them. The IR sensor provides a fast, simple, reliable trigger, and the camera plays the role of investigator trying to understand what really happened. Automation stops being a pile of random boxes from different vendors and turns into a clear architecture: simple sensors maintain the baseline of common sense, and computer vision kicks in when life once again reminds us that the world is a bit more chaotic than the diagram in the equipment manual.