Different from the conventional frame-based image sensor, the dynamic vision sensor (DVS) is an event-based image sensor. The DVS only records the pixels which have light intensity change in time and results in much lesser data to be processed than the conventional image sensor. After processing the changed pixels, DVS can transmit the pixel information to the back-end processor immediately. The overall process is fast with low latency. The DVS can operate under a wide range of ambient light conditions with its event-based feature, since the sensor is only sensitive to the change of ambient light intensity rather than the intensity itself. The lower data volume of DVS can lead to lower power consumption and enable power saving application. General speaking, the DVS devices generate lower data volume, ultra-fast response time, high dynamic range and low-power consumption, which is suitable for production line inspection, environmental perception of unmanned systems, people counting in retail stores, driver and occupancy monitoring in vehicles, etc. Furthermore, the output of DVS has no color information, no detailed features of people or objects, there will be no privacy issue. Thus DVS is very suitable for detection of accidental events in hospitals and homecare.
The following figure shows the difference between the output data generated by the DVS and by the standard camera.
The conventional image sensor outputs all pixel values in a frame during each frame sampling interval. On the other hand, the DVS only outputs the coordinate of pixels where their light intensity has changed, thus the amount of output data can be greatly reduced. The following figures show the difference between a DVS image frame and a frame generated by the conventional image sensor.
1. Which DVS suppliers and products are there in the market?
Companies currently developing dynamic vision sensors include: iniVation, Samsung, Prophesee, CelePixel (acquired by OmniVision in 2020), Insightness (acquired by Sony in 2019), etc. The resolution of DVS devices are mostly between 300,000 to 1.2 million pixels. The output interface of DVS is either a parallel bus or a mobile industry processor interface (MIPI) serial link. The transmission control protocol of DVS is mostly based on the address event representation (AER) with many vendor dependent customization. The application processor or microprocessor needs to connect with these sensors through an FPGA which implements the corresponding event decoder. Therefore, the product designed with the dynamic vision sensor becomes more complicated and costly than with the conventional sensors.
The current DVS camera solutions available in the market, such as iniVation’s DVXplorer and Prophesee’s EVALUATION KIT2-HD, are all development platforms with diversified interfaces so that developers can connect it to the back-end system to realize new concept designs or develop new prototypes. However, they are not suitable for use as a component in the end product development. As the first home surveillance camera based on an in-house commercial DVS component, Samsung’s SmartThings Vision  emphasizes privacy protection and human detection for smart home security monitoring and accident detection, such as fall-down detection and intruder detection. Prophesee and Xperi  cooperated to develop the driver and occupancy monitoring solutions with DVS. Through the DVS sensor’s advantages of fast response speed, high dynamic range and low data volume, they’ve implemented algorithms such as eye tracking, head posture detection, glasses/mask detection, and eye opening or closing detection.
2. Which are the markets for DVS applications? How is the estimated future market size?
Since the photosensitive operation principle of DVS is similar to that of the photosensitive neurons of animals, DVS is considered to be a neuromorphic vision sensor. It has been applied in many different fields, such as industrial automation, autonomous driving, smart home and security monitoring, etc. The DVS sensor imitates the characteristics of the human retina and transmits data only when the pixel brightness changes. It can reach the microsecond level of response time, and can also greatly reduce the amount of transmitted data to lower the energy consumption and minimize the computation power needed for data analysis. This sensor can also support high dynamic range imaging in high-contrast lighting conditions such under the bright sunlight and during the night.
According to Yole Development's market analysis, by 2029, the neuromorphic semiconductor market for sensing and computing can reach 7.1 billion U.S. dollars. However, there are still technical problems waiting to be overcome. If they can be resolved in the next 4 to 5 years, the neuromorphic computing market is projected to reach USD 69 million in 2024, and will grow to USD 5 billion in 2029 and reach USD 21.3 billion in 2034. The market size of the neuromorphic sensing in 2024 is estimated to be USD 34 million, and it will grow to 2 billion U.S. dollars in 2029 and to 4.7 billion U.S. dollars in 2034.
3. What are the current problems for DVS applications?
At present, there are not many DVS-based end products available in the market. There are several reasons. The first reason is because the price of DVS is much higher than the traditional image sensor with the same resolution, the product developers thus hesitate to use it due to higher product BOM cost. The expensive end product price also discourages the consumer acceptance of such DVS-based products. This further reduces the willingness of manufacturers to develop new products with DVS. On the other hand, In the areas of industry, commerce, and infrastructure, solving problem is more important than the sensor price, so DVS application is known to the machine vision market. The second reason is that there is almost no application processor or image processor that can directly support DVS. FPGA is usually used to process the event output of DVS devices, which increases the difficulty and cost of product design. The third reason is that the computer vision and deep learning algorithms suitable for event data processing are not common. This further drives the manufacturers away from developing DVS applications.
In response to the above problems, DVS manufacturers have planned to control the cost of DVS sensors to be close to that of traditional image sensors, so that the product manufacturer can develop products with acceptable price for the end customers. In addition, some IC developers have already designed application processor or image processing chips that can support DVS data access protocol as well as various other type of sensors. This helps the developers to simplify the use of DVS components and encourage them to devote resource and time to the end product development, and further reduces costs and accelerates time to market. As a result, the algorithm developer can obtain the DVS development platform easily and at a low price. The demand of various applications will also appear. Consequently, algorithm developer will further invest in the development of DVS-based solutions for more end product applications.
From the perspective of competition, if we can establish DVS application technology, create our own algorithms, and solve the difficulties of end consumers, there is a chance to build certain barriers in the early stage and then extend to the other product areas to become a leader in that field.
In the current consumer market, Samsung SmartThings Vision home camera is the most known DVS application product. The other manufacturers’ products are not widely award. Based on the unique advantages of DVS such as fast response time, low power consumption, and low data transmission rate, high dynamic range, high privacy protection, etc., DVS is suitable for use in machine vision related fields such as: industrial control, commerce, retail, autonomous driving, smart home, security surveillance, etc. It's believed that there will be more manufacturers and academic institutions who will enter the DVS market in the future, including sensor suppliers, algorithm suppliers, application processor providers, image processor developers and application developers. They can work together to solve problems and accelerate the application of DVS to create a new machine visual application market.
Author Edward Kuo