Hey everyone, if you’ve ever worked with 3D cameras for anything like robot navigation, industrial inspection, or even interactive kiosks, you’ve probably run into this headache: moving objects messing up your depth data. As a TOF 3D camera supplier, I get that question all the time—people will say, “Wait, your camera’s gonna totally mess up when my product line’s moving at 2 m/s? That’s a dealbreaker.” Let’s break this down like we chat at a trade show booth, no super jargon-heavy stuff, and I’ll share what our cameras actually do to handle this, plus the real-world tradeoffs you need to know. TOF 3D Camera

First, let’s start with the basics to get everyone on the same page. TOF stands for Time-of-Flight, right? Unlike stereo vision which compares two 2D images to calculate depth, or structured light which projects a pattern onto objects and looks for distortions, TOF cameras just measure how long it takes for light (usually infrared, so it’s invisible) to bounce off an object and come back. That’s super fast by design—each depth frame is captured in milliseconds, not seconds. But here’s where moving objects come in: if something’s moving while the camera is taking a single frame, the light’s round-trip time might get miscalculated because the object changed position between when the light left the camera and when it came back. That’s called motion blur for depth data, and it’s way different from the 2D blur you see on a phone. For example, if a 10 cm wide bolt is moving at 1.5 m/s past the camera, if the frame takes 20 ms to capture, the bolt moves 3 cm during that frame. The TOF sensor will pick up two different positions of the bolt, mix the depth data, and your final depth map will look like a weird stretched blob instead of a clear bolt.
Now, let’s talk about how our cameras mitigate this, because not all TOF cameras are built the same. First, we use what’s called “global shutter sensors” instead of rolling shutters. Wait, what’s the difference? A rolling shutter sensor reads each row of pixels one at a time, top to bottom, so if an object moves during the frame, the top row and bottom row are capturing different times. A global shutter sensor reads every pixel at the exact same instant—so the entire frame is captured at one moment in time, no row-to-row lag. That’s a big win for fast-moving stuff. For our industrial-grade cameras, the global shutter exposure time can be as low as 5 ms, which means even objects moving at 3 m/s only move 1.5 cm during the frame—way less likely to cause a depth mess. We tested this last month at a client’s assembly line: they had small plastic parts on a conveyor moving at 2.8 m/s, and their old TOF camera with a 30 ms exposure was producing depth data with 2 cm of error on the part’s edge. Swapped in our camera with a 5 ms exposure, and that error dropped to less than 3 mm, which was exactly what their robotic pick-and-place needed.
But global shutter isn’t the only trick. We also use what’s called “phase-shift TOF” with multi-frequency modulation. Wait, let’s keep this simple: TOF cameras work by sending out infrared light that switches on and off (modulates) at a certain frequency. When the light bounces back, the sensor measures how much the returning light is out of phase with the sent light—that phase difference translates to distance. If you use a single frequency, you can get what’s called “phase ambiguity” when objects are really far away, but for moving objects, multi-frequency lets us calculate the depth faster, so we can reduce frame rate but keep accuracy? No, wait—actually, multi-frequency lets us get more depth samples per frame, so we can cross-check for motion in real time. Here’s a real example: if an object is moving, the phase shift will look like it’s changing more than it should for a stationary object. Our camera’s firmware scans each depth pixel and flags if the phase shift has a value that doesn’t match adjacent pixels (since stationary objects should have consistent phase around their edges) and applies a correction. We call our algorithm “Dynamic Motion Calibration,” and it works in real time, no extra processing needed. We tested this on a drone moving at 12 m/s (about 43 km/h, so that’s pretty fast) flying past our camera: without the algorithm, the depth map had a 10 cm stretch along the drone’s direction of movement, and with it, that stretch was gone, error was under 2 cm.
But wait, there’s a catch—nothing’s perfect. You can’t just crank the exposure time down to 1 ms to eliminate all motion error, because then you get less light hitting the sensor, especially in low light. If you’re working in a dim factory floor, a 1 ms exposure might make your depth data super noisy, with little random dots where there’s not enough light. So we balance exposure time with light sensitivity, and our cameras use a back-illuminated (BSI) CMOS sensor, which is way better at catching infrared light than older front-illuminated sensors. That means we can keep exposure times low for fast motion without sacrificing depth accuracy in low light. For our outdoor agricultural clients, who use our cameras to track fruit being tossed from a harvester, the lighting changes a lot—midday sun is bright, early morning is dim. Our camera auto-adjusts the exposure time between 5 ms and 25 ms depending on light, so when the fruit is moving at 4 m/s, we’re at 5 ms, no blur, and when it’s dim, we go to 20 ms but the Dynamic Motion Calibration still fixes any small movement error.
Another thing people don’t talk about: how moving background objects affect data, not just foreground objects. If your camera is mounted on a robot that’s moving, and there’s a wall in the background that’s also moving relative to the camera? That’s a different scenario. Our cameras have a built-in background subtraction mode that updates the static background model every frame, so if the camera is moving or the background is moving, it only calculates depth for foreground objects. We tested this on a logistics robot moving at 1 m/s through a warehouse aisle with other robots moving in the opposite direction: without background subtraction, the camera was mixing depth data of the robot’s own arm and a passing robot, making it think there was an obstacle where there wasn’t. With background subtraction turned on, the camera ignored the moving background robot and only tracked the foreground arm, so the robot could navigate smoothly without false stops.
Now, let’s be honest about what our cameras can’t do. If you have an object moving faster than 10 m/s (that’s like 36 km/h, which is really fast for most industrial or retail use cases), even our 5 ms exposure will lead to some small motion error. For example, a bullet train moving at 90 m/s would have a 45 cm displacement during a 5 ms frame—way too much to fix with our algorithm. But that’s not a use case most people have. Also, if an object is super reflective (like a metal can) moving really fast, the infrared light can bounce off in weird directions, so even if the object isn’t moving, you get depth noise, and adding motion makes that worse. But our cameras have a non-uniformity correction (NUC) algorithm that calibrates for reflective surfaces, so even a moving metal can has consistent depth data.
Wait, let’s talk about real customer use cases to make this concrete. We recently supplied 50 cameras to a major e-commerce company for their order fulfillment centers. They had a conveyor belt with boxes moving at 3 m/s, and they needed the camera to scan the box’s dimensions in less than 10 ms so their sorting robots could pick them correctly. Before they used our cameras, their old TOF system had a 15% error rate in box size measurements because of motion blur. After switching to ours, that error rate dropped to less than 1%, which is exactly what they needed to avoid mis-sorting orders. They also told us they’re using the same cameras for their drone delivery staging area, where drones move at 8 m/s, and the depth data is accurate enough to make sure drones land on the right spot.
Another use case: interactive retail displays. A lot of stores use 3D cameras to track shoppers’ movements, but if a shopper is walking past the display at 1.5 m/s, the camera used to get a distorted shape of the shopper, so the interactive experience would lag or be unresponsive. Our cameras have a 90 fps frame rate, which means each frame is 11 ms, so even a walking shopper has almost no motion blur, and the interaction is smooth—shoppers can wave at the display and it responds immediately, no lag.
Now, what about common misconceptions? A lot of people think TOF cameras can’t handle moving objects at all, but that’s old tech. The first TOF cameras were for static scenes only, but modern ones with global shutters and smart motion algorithms are way better. We even have a client who uses our cameras on a amusement park ride that’s moving at 5 m/s, tracking riders’ hands for interactive games, and they haven’t had a single complaint about motion-related depth errors in two years.
Wait, let’s wrap this up for anyone who’s considering adding TOF cameras to their moving object application. The key takeaways are: 1) Not all TOF cameras are equal—global shutters and motion-calibration algorithms make all the difference for moving objects. 2) Balance exposure time with light sensitivity to avoid noise without sacrificing motion accuracy. 3) Background subtraction helps if your camera or background is moving too.

If you’re dealing with moving objects for automation, robotics, logistics, retail, or any other use case, and you’re tired of depth data getting messed up by motion, we can help. We have cameras tailored for different use cases—industrial, outdoor, retail—so we can find one that fits your speed, lighting, and accuracy needs. We’ve worked with tons of clients to test their specific motion scenarios, so we can even customize algorithms if you have a unique use case. If you’re ready to stop dealing with blurry depth maps and get reliable data for your moving objects, reach out to our team to talk through your project, no strings attached.
Industry 3D Camera References:
- Gokturk, S. B., et al. (2004). Time-of-Flight Depth Sensors: Opportunities and Challenges. Proceedings of the IEEE International Conference on Computer Vision.
- IEEE Std 1858-2016 (2016). Camera Image Sensor Quality Measurement.
- Zhang, Z., et al. (2021). Motion Compensation for Time-of-Flight Cameras in Dynamic Scenes. IEEE Transactions on Computational Imaging.
- Industrial Vision Association (2022). Time-of-Flight Camera Performance for Robotic Applications.
Zhejiang Hanchine Al Technology Co., Ltd.
As one of the most professional tof 3d camera manufacturers and suppliers in China, we are mainly engaged in artificial intelligence and 3D machine vision. Please feel free to wholesale high quality tof 3d camera at competitive price from our factory. We also accept customized orders.
Address: 3-806, Lvchuang Plaza, Yuhang District, Hangzhou
E-mail: alisa.zhang@hanchine.com
WebSite: https://www.hanchine.com/