Design Article
How to implement automotive smart rear-view cameras
By Tom Wilson, CogniVue Corp.
1/21/2013 1:55 PM EST
Developing algorithms
Vision algorithms are developed on the PC using established computer vision frameworks such as Matlab and OpenCV (an open source library of functions aimed at vision applications). During the algorithm development phase, the developer should be mindful of the architecture, properties and limitations of the embedded platform in order to ease optimization and porting processes.
If the embedded target software development kit (SDK) also supports a PC platform version, it is preferable to use these data structures and API whenever possible to ensure a smoother porting of the algorithm. Once the PC algorithm is validated and performs satisfactorily, it is optimized based on the properties of the embedded platform. If the optimized PC algorithm adheres to the resource and power limitations of the embedded platform, the developer proceeds to port the algorithm to the target device. If not, a minimized or ‘scaled-down’ variant of the algorithm is developed and ported to the chip.
The minimized variant uses approximations simplifying assumptions and substitutes algorithmic functions with ‘lighter’ versions to produce a vision algorithm that approximates the performance of the original PC algorithm. This minimized algorithm is much less computationally and resource intensive than the original PC version. Once the ported algorithm is validated on-chip, field trials start. Running field trials augments the validation testcases and tests the performance in real-world scenarios.
The following figure illustrates typical algorithm optimization stages:
Requirements of the Smart Rear-view Camera System
The system requirements as dictated by the end-customer – an after-market global supplier of automotive parking assist systems – were challenging. The camera module needed to be better quality than existing rear camera systems and have a bill of materials less than $35 to ensure an attractive average selling price. The product specification required that it be capable of real-time (>25 frames per second) de-warp and perspective correction, object detection, tracking and distance estimation of the closest obstacle to the bumper. The customer’s intent for estimating distance was to replicate the ‘zones’ used in ultra-sonic parking sensors. Also, the system had to support multiple viewing modes including panoramic, bird’s eye and left/right split views and render graphic overlays to the display.
Next: Customizable code
Vision algorithms are developed on the PC using established computer vision frameworks such as Matlab and OpenCV (an open source library of functions aimed at vision applications). During the algorithm development phase, the developer should be mindful of the architecture, properties and limitations of the embedded platform in order to ease optimization and porting processes.
If the embedded target software development kit (SDK) also supports a PC platform version, it is preferable to use these data structures and API whenever possible to ensure a smoother porting of the algorithm. Once the PC algorithm is validated and performs satisfactorily, it is optimized based on the properties of the embedded platform. If the optimized PC algorithm adheres to the resource and power limitations of the embedded platform, the developer proceeds to port the algorithm to the target device. If not, a minimized or ‘scaled-down’ variant of the algorithm is developed and ported to the chip.
The minimized variant uses approximations simplifying assumptions and substitutes algorithmic functions with ‘lighter’ versions to produce a vision algorithm that approximates the performance of the original PC algorithm. This minimized algorithm is much less computationally and resource intensive than the original PC version. Once the ported algorithm is validated on-chip, field trials start. Running field trials augments the validation testcases and tests the performance in real-world scenarios.
The following figure illustrates typical algorithm optimization stages:
Figure 2: Algorithm Optimization Steps
Requirements of the Smart Rear-view Camera System
The system requirements as dictated by the end-customer – an after-market global supplier of automotive parking assist systems – were challenging. The camera module needed to be better quality than existing rear camera systems and have a bill of materials less than $35 to ensure an attractive average selling price. The product specification required that it be capable of real-time (>25 frames per second) de-warp and perspective correction, object detection, tracking and distance estimation of the closest obstacle to the bumper. The customer’s intent for estimating distance was to replicate the ‘zones’ used in ultra-sonic parking sensors. Also, the system had to support multiple viewing modes including panoramic, bird’s eye and left/right split views and render graphic overlays to the display.
Next: Customizable code
|
|
|
|
|
Navigate to related information
William Miller
3/22/2013 8:34 AM EDT
I support every single decision to make our lives safer. If a back-up camera can save someone's life, why didn't put it in your vehicle? Trucks and SUV should be the first in the queue I guess.
_________________
William - http://www.carid.com/back-up-cameras-sensors.html
Sign in to Reply