Named as one of CB Insights
100 Most Promising AI Startups 2023
By Oren Debbi CEO And Co-Founder, Visionary.Ai
Software-based image processing powered by AI chips will be used by applications to improve road safety, guide surgeons and enable inspection of public infrastructure and crops, plus hundreds of additional use cases, says Oren Debbi, CEO, and co-founder of Visionary.ai
Improvements in image processing have enabled major advances in several industries. The sensor industry is expected to reach 36 billionby 2027, which is expected to grow by 8% yearly. These sensors power hundreds of applications, from improving road safety to monitoring endangered species.
While these sensors can produce clear and accurate images when the lighting is good, an estimated 85 % of devices operate in conditions where light levels are poor. This can be caused by heavy rain, fog, High Dynamic Range (for example, a person silhouetted in front of bright background), high motion, and sudden changes in intensity, such as flashing headlights.
In the coming year, software-based image processing powered by highly advanced AI chips will advance to improve image quality further, even in the most challenging conditions. Below are just a few examples.
When performing laparoscopic surgery, a surgeon uses a probe or endoscope with a built-in camera to explore internal areas of the body, and the resulting image is displayed in real-time on the computer monitor. This procedure diagnoses illnesses such as cancer and liver disease, removes a damaged organ, or takes a tissue sample for further testing (biopsy).
Several challenges stand in producing crisp, clear streaming video. To make surgery less intrusive, the sensor and LED lights need to be small to fit on the tip of the endoscope, and the amount of light that is used needs to be limited since tissues have liquids and fats which are highly reflective.
In 2023, a software-based ISP will separate the signal from the noise providing a superior, clear image at lower illumination levels while keeping the sensor size to a minimum. This software-based ISP will be customized to focus on red and brown shades without reading in colors that don’t exist in the human body. Distortions will be eliminated, enabling surgeons to see every fine detail down to the blood vessels. Since the algorithm to enhance the video will be AI-based deep learning will be applied to improve image quality continuously.
There are many applications designed to make travel more comfortable and safer. Sensors used to power these applications must deliver excellent performance in all visibility conditions, even when there is glare from wet roads and rain, fog, or snow that can limit visibility.
Next year, an advanced denoiser can use the image sensor’s raw output data, which has not been compressed or degraded by any post-processing, to improve image clarity. The performance will be improved by working alongside a software image signal processor to create a solid foundation for an evolving suite of image enhancement features that can improve driver safety.
Car makers and safety regulators already have a strong interest in driver alertness monitoring. The time that tiredness is nighttime when the cabin is most dark. Clear images of the driver’s face and eyes, using extremely low ambient light, will enable algorithms to assess the attention or drowsiness of the driver accurately.
See More: Sensors and Sensibility: How to Start Your AIoT Program for Best Results
Today drones are being used to plant seeds and spray crops to save farmers time and money invested in labor and equipment. Next year by using smart cameras powered by video enhancement, farmers can detect pests in real time and effectively take action against them without necessarily harming helpful insects.
These drones will also use enhanced video images to detect the locations of weeds with a high level of precision to determine where to apply herbicides. Farmers will acquire more accurate data to apply the minimum amount of chemicals necessary by using advanced denoisers matched with a software-based image processor.
For livestock farmers, drones equipped with cameras will also be used to pick out potentially sick or diseased animals. Drones will use sensors to measure vital information like size and physical activity. Machine learning and facial recognition systems will identify each animal. All of these actions will be improved due to enhanced video processing.
Due to the ability to operate in low light, drones can also fly more hours each day, making precision farming more economical.
Parks, highways, bridges, and roads are regularly inspected to protect public safety. Taking video footage regularly provides project officials with confidence that any new work has been done according to specifications and that structures are well maintained and don’t pose any hazards like sinkholes or dangling electrical lines.
In the coming year, drones equipped with cameras will be used to inspect hard-to-reach places in hilly or mountainous terrain, checking for roof damage, pipeline leak detection, and wind turbine inspections so that the issues can be detected and repaired as soon as possible.
Advanced software-based image processing will enable drones to detect small cracks and leaks that might not otherwise be visible. In addition, the ability to operate in low light conditions will enable drones to fly at dawn and twilight to conduct more daily inspections. Since detecting any damage and responding quickly is important to prevent further damage that could disrupt transportation and threaten lives, having more accurate information faster will significantly boost public safety.
In 2023 we expect that advanced software-based imaging processing will provide more accurate and up-to-date video to improve health, safety on the roads, agricultural yields, and the reliability of public infrastructure. More accurate sensors that can relay information in real time will improve the effectiveness of mission-critical applications that can save lives. As AI algorithms continue to improve performance and more specialized chips are launched, the accuracy and speed of software-based imaging processing will continuously evolve, with new capabilities being discovered every day.
Read the full article here.