Axis Camera Application Platform (ACAP) LaunchedBy: John Honovich, Published on Oct 18, 2009
Axis has announced an open application platform for video analytics. This allows customers to run a variety of 3rd party video analytics on Axis cameras. Simultaneously, video analytic manufacturers gain access to Axis' extensive user base. The first applications are planned for Q4 2009 with AgentVi, Aimetis and Via:sys among the early adopters. [Note: see Axis section on their website describing their Camera Application Platform.]
Limitations of camera and recorders support is a historical challenge for the adoption of video analytics. Some video analytic manufacturers only support their own brand of cameras, while most camera manufacturers offer no video analytics or a single option for their cameras.
The most optimistic analogy is to Apple's iPhone app store which has single-handedly exploded the use of applications on mobile phones.
While Axis's program should remove limitations on camera support, the most significant concern will be the capacity and performance of the analytics running on Axis cameras.
To that end, we see 3 key questions that will determine the success of Axis analytics program:
- How well will the analytics perform on Axis cameras?
- How will performance vary on different Axis cameras?
- How significant will support and optimization of analytics be on Axis cameras?
From a user/integrator perspective, 3rd party video analytics software will be uploaded to the cameras in the same manner firmware upgrades are executed.
From a video analytics manufacturer perspective, the developer ports their C/C++ code to Axis cameras (running Linux) using the Axis Embedded Developer SDK (see the Axis developer Q&A). Companies must register for the SDK (see registration form).
Running on Axis Cameras
Video analytics will be run on Axis's chipset. However, H.264 encoding and image processing (the most resource intensive elements) are performed in hardware and do not require CPU usage. As such, Axis estimates that the CPU is lightly utilized and available for performing video analytics.
The amount of computing resource available depends on chipsets used. The current generation chipset is the ARTPEC-3 [link no longer available] which is used in most of the newly introduced products such as the Q1755 and the P33 series cameras. The next generation chipset, the ARTPEC-4, is planned for release in 2010. This chipset will be substantially more powerful but specific details have not been released. However, many of Axis top selling cameras use the older generation ARTPEC-2 chips (for example, the 211, 221, 223M). The analytics platform will not be offered for ARTPEC-2 cameras (though some analytics providers such as AgentVi and Via:sys currently run on them)
Examining the current generation ARTPEC-3 chipset raises some concerns on performance.
Axis describes the ARTPEC-3 as a SOC that "integrates a powerful CPU, an Ethernet (10/100) controller, MMU, and a full L1/L2 cache. Image and video encoding to JPEG, Motion-JPEG, and H.264, and high quality image processing, are supported in dedicated hardware."
Hardware specifications of the ARTPEC-3 include:
- 200 MHz RISC CPU with a 32-bit data and address width
- Separate instruction and data (L1) caches, each of 16 kB and 2-way set associative
- 64 kB Level-2 cache, 8-way set associative
- Memory controllers for DDR2-400 external RAM
Aimetis reports that, "With the ARTPEC-3, we will have some of our retail analytics available. With ARTPEC-4, there is enough CPU for all of our algorithms except our Auto-PTZ tracking algorithm. "
A review of these high-level specifications with 3 independent video analytic manufacturers raised some concerns about the relative processing power compared to current chip offerings used for video analytics. Specific chips cited were the TI 6446 [link no longer available] and the TI 6437 [link no longer available]. While they could not make a definitive conclusion with the limited information provided, they were concerned about the computing resources available in the current ARTPEC-3 chip.
3rd Parties Responsible for Ensuring Compatibility
Each video analytics manufacturer will be responsible for testing and ensuring that their analytics run on various Axis cameras without compromising the camera's performance. Axis will provide the manufacturers a compatibility tool. The guarantee will be the responsibility of the analytics manufacturer, not Axis.
A copy protection tool can be enabled to ensure proper licensing. According to Axis, "The analytics vendor requests License Codes for their application from Axis. The vendor sell and distribute these License Codes to their customers/integrators. To unlock the application, the License Code together with a camera-unique ID is used to retrieve a License Key from Axis website. This License Key is uploaded to the camera and the application is unlocked. The vendor can provide also a Trial Period using the same scheme."
Support for configuration and optimization of the video analytics is the sole responsibility of the video analytics provider.
Axis is not charging a licensing fee or requiring revenue sharing. This is essentially free to the video analytics manufacturer (who, of course, charges a license fee to the customer).
Axis anticipates profiting from this platform from the growth of camera sales.
Issues to Consider
While the openness of the platform, combined with Axis large size, with draw significant interest, there are a number of issues to consider:
- Potential insufficient computing power to achieve high performance: Computing requirements vary greatly based on the type of analytic performed, the sophistication of the analytic, the efficiency of the implementation, etc. Many of the "higher end" analytics may not be able to run on the ARTPEC-3 cameras. This may be resolved as cameras using the upcoming ARTPEC-4 are released. However, this is a key unknown
- Challenges in determining what cameras support what analytics: Because Axis supports a broad range of cameras using multiple generations of chipsets, it could be somewhat challenging to determine which cameras can support the analytics required.
- Service issues due to analytics overload: Since analytic vendors are responsible for verifying their own conformance, this could lead to problems in the field (at least initially until the issues are identified and notified to Axis, etc.)
- General performance of analytics: As with all analytics, the performance and need for on-site optimization will still be a key concern. Given that these analytics will be running on a large number of models and the analytics manufacturer may have little field experience with these models, model specific issues may arise.
- ObjectVideo - Axis's platform may be complimentary to ObjectVideo yet has the potential to undermine OV's business strategy. ObjectVideo aims to be the "Intel Inside" of video analytics, providing video analytic software that powers video surveillance systems for manufacturers across the board. To this end, ObjectVideo could run on the Axis platform just like it runs on TI and Intel processors. On the other hand, to the extent that the Axis platform is successful, it has the potential to enable smaller video analytic startups to enter the market and leverage the strength of Axis's market share.
- ioimage [link no longer available] - Axis approach is probably most significantly different than ioimage's. ioimage runs its analytics only on its own cameras and encoders with the goal of maximizing performance and quality. To the extent that Axis approach can provide similar performance in an open platform, this would undermine the value of ioimage's more 'closed' and tightly integrated approach.
- Small video analytic companies may benefit the most from the success of Axis's analytic platform. Today, there are dozens of small video analytic developers world-wide who generally run on servers or have OEM agreements with IP camera manufacturers with small market share. The Axis platform could make it much easier for users to deploy analytics from smaller companies.
- Bosch - Bosch has a similar approach architecturally to ioimage, integrating their own analytics in their cameras and encoders. The risk to Bosch is similar to that of ioimage. Of course, since Bosch has a significantly larger market size than ioimage, the risk is not as important to their overall business.
- ACTi / budget camera companies - None of the 'budget camera' companies run high-end on board analytics. Axis's analytic platform provides a differentiator for mid to higher end projects that seek to use video analytics. On the other hand, the more powerful chips required to run analytics may maintain or extend the cost advantage the budget camera companies offer.
- Milestone's analytics platform - Milestone offers a similar open analytics platform [link no longer available] in which dozens of vendor's analytics can be integrated and monitored through Milestone's VMS. The Milestone platform is complimentary to the Axis platform. Customers may run video analytics on Axis cameras and then have the analytics monitored on Milestone VMS software. The Axis platform is likely to have a bigger impact on the overall market due to Axis's much larger market size and the flexibility of adding a few cameras with video analytics to any project.
- VMS companies - This platform does not compete with VMS providers. Indeed, since VMS providers often develop their own video analytics, the VMS providers may see this as an opportunity to off-load some processing power from their servers (for example, advanced motion detection, facial detection, people counting, etc.). They may offer this for free (or low cost) as an incentive to use their systems.
Long Term Outlook
Some important questions exist on the performance of analytics running on Axis cameras.
The key question is likely "when" rather than "if". Will this be mass market in 2011 or 2013, etc? Even if some video analytics require too much processing power for the current generation, at some point in the near future, the cameras likely will be capable of doing do and at an incremental hardware cost that is minimal.