ANALYSIS: EMBEDDED VISION
Data processing: The sheer volume of data generated by multiple high-resolution cameras can overwhelm processors and memory. Efficient data handling and processing algorithms are required to avoid bottlenecks.
Calibration: Calibrating multiple cameras to ensure they produce accurate and consistent images is a complex task. Factors such as lens distortion, colour variation, and geometric alignment must be carefully calibrated for reliable results.
Multi-camera set-ups can be used with ‘smart trolleys’ to deliver an autonomous shopping experience
impacting system design and functionality. But the more difficult part is to carefully evaluate and consider the following factors when integrating multi-camera systems.
Four critical factors to consider while integrating multiple cameras
Number of cameras Determining the appropriate number of cameras is crucial. Factors include the required level of precision, the characteristics of the target object, the desired field of view, and the processing capabilities of your host platform. More cameras can offer several benefits, including improved resolution, reduced lens distortion, and broader coverage.
Method of synchronisation Two synchronisation methods should be on your radar. First, there’s software synchronisation, which proves effective when capturing static objects within controlled environments. This method relies on software algorithms to align captured frames. It provides a practical solution for situations where pinpoint synchronisation isn’t a major requirement. However, hardware synchronisation
is the go-to choice for scenarios involving moving objects and stringent synchronisation demands. It involves initiating simultaneous image capture on all cameras through a hardware trigger, typically an external pulse-width modulation (PWM) signal. This ensures perfectly aligned frames, making it ideal for embedded vision applications that demand real-time decision-making capabilities.
Camera interface Your bandwidth requirements will predominantly dictate the choice of
camera interface. For applications where high-resolution images need to be captured at rapid frame rates, opting for a MIPI interface is often favoured over USB. Consider factors such as data transfer distance, reliability, and compatibility with your host platform. When dealing with high-bandwidth data transfer over extended distances (beyond 2-3 metres), exploring interfaces such as GMSL2 or FPD-Link III becomes necessary.
Host platform The host platform is an important component of your multi-camera system. Various processors are available on the market, each with their own strengths. The NVIDIA Jetson series stands out as a popular and advanced choice, but alternatives such as the NXP
i.MX series, Qualcomm, and Texas Instruments are also worth considering. Your choice should hinge on multiple
factors, including processing power, form factor, cost constraints, AI performance measured in TOPS (tera operations per second), power consumption, thermal performance, maximum camera support, required interfaces, and compatibility with the software ecosystem that you plan to use.
Multi-camera integration challenges While integrating multiple cameras into embedded vision applications can offer amazing benefits, it also comes with its fair share of challenges. These include:
Synchronisation: In applications where timing is necessary, confirming that all cameras capture images simultaneously can be technically demanding. Even a slight delay between cameras can lead to inaccuracies in data capture and analysis.
22 IMAGING AND MACHINE VISION EUROPE DECEMBER 2023/JANUARY 2024
Bandwidth and interfaces: Making sure the chosen camera interfaces can handle the data bandwidth is crucial. High- resolution cameras at high frame rates will require interfaces that can transfer data without loss or latency.
Power and heat management: Operating multiple cameras simultaneously can lead to increased power consumption and heat generation.
Cost: The cost of multiple cameras, interfaces, and processing hardware can add up very quickly.
Field of view: Aligning the field of view of multiple cameras to cover the desired area can be tricky. Variations in camera placement and orientation can lead to blind spots or overlapping areas, requiring careful planning and adjustment.
Environmental factors: Embedded vision applications often operate in diverse environments, presenting challenges such as variable lighting conditions, weather, and vibrations. These factors may impact the performance of cameras and the accuracy of vision algorithms.
Maintenance: Securing the long-term reliability of a multi-camera system can be tough. After all, cameras and components may require periodic calibration, cleaning, or replacement, and addressing issues in embedded systems can be complex. While these challenges may seem like
long, difficult paths, the truth is that you can easily cross them by partnering with the right embedded vision expert. But before you do, it’s highly recommended that you understand what these factors entail and how they collectively determine the effectiveness of your multi-camera system-based application. After all, multi-camera systems are likely to directly impact the future of various industries, changing how humans and machines capture, analyse, and interact with visual data. I
@imveurope |
www.imveurope.com
e-con Systems
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32
