Computer architecture for vision systems
- 1. Computer Architecture for Vision Systems Presented by: Ritesh Thakur ( 30 ) Vishal Thoke ( 32 ) Utsav Patel ( 36 ) Sushant Vaidkar ( 38 ) Vedant Valsangkar ( 40 ) Bansilal Ramnath Agarwal Charitable Trust's Vishwakarma Institute of Technology
- 2. The Evolution of Vision Technology Computer vision:research and fundamental technology for extracting meaning from images Machine vision:factory applications Embedded vision:thousands of applications 1)Consumer, automotive, medical, defense, retail, gaming, security, education, transportation, … 2)Embedded systems, mobile devices, PCs and the cloud
- 3. What is Computer Vision ? • Out of the five senses, Vision is known to be the superior source of data in human beings. We can't do our daily tasks without our eyes. • Eyes are the major source of information. What if computers can also analyse visual data using Cameras? • Computer Vision is an interdisciplinary scientific field that deals with how computers can gain high level understanding from digital sources like images and videos.
- 4. Every Computer Vision System Looks Something Like This Camera Local Processor Network Connection Cloud Backend
- 5. ARCHITECTURE Analyzing the process of a computer vision application, one can perceive that the different tasks demand different efforts from the various computational resources. One can conclude that a single processor architecture may not be able to carry all these operations efficiently; there is a need for a hybrid processing configuration, with specific architectures for each level.
- 6. Architectures for low-level operations are heavily explored, due to their specific characteristics and the large amount of data involved. There is a trend to use SIMD (single instruction multiple data) parallel architectures for low-level processing, and a second architecture for medium and high-level operations. Digital signal processors (DSPs) have also been used for low level operations, with good performance
- 7. Both consist of pixels which use the Photoelectric effect to generate electric signals Types of Camera Sensors: CCD (Charged Coupled Device) ● Passive-pixel device ● Less noise in pixel data ● Expensive, requires more power ● Used in high quality video cameras and satellites CMOS (Complementary Metal Oxide Semiconductor) ● Active-pixel device ● More noise ● Affordable, low power consumption ● Used in smartphones and DSLR
- 8. Types of Camera Sensors: CCD (Charged Coupled Device) ● No electronics at pixel level ● All electric signals need to be transferred to external electronics for conversion into voltage ● Hence, the sensor is quite slow ● Provides better quantum efficiency ● Ideal for poor lighting conditions CMOS (Complementary Metal Oxide Semiconductor) ● Each pixel contains separate electronics e.g. amplifier ● The signal from each pixel can be read directly without any changes ● Provides higher frame rate ● Image is scanned row-wise causing rolling shutter effect
- 9. Applications of Computer Vision Technology 1. Automotive Safety : ● Vision system can assure safety of vehicles in auto pilot mode. ● Using cameras we can detect objects nearby and can avoid the obstacles.
- 10. Applications of Computer Vision Technology 2. Tracking Objects : ● Using surveillance cameras we can keep track of household items. ● We can keep an eye on our important stuff.
- 11. Applications of Computer Vision Technology 3. Hazardous Areas Scanning : ● Cameras and drones can go where human eyes can not reach. ● Humans can take a look inside the hazardous areas using drones.
- 12. Applications of Computer Vision Technology 4. Biological Applications : ● Small cameras are used in surgeries to detect the area of body. ● Vision systems can also detect various samples of microbes and DNA.
- 13. Development: Future •Heterogeneity of hardware becomes hidden •OpenVX: Abstracts hardware, not the algorithm •Higher-level APIs: Abstract the algorithm and hardware •Higher-level deep learning abstractions •Automated optimization of neural networks •Automated design and training of neural networks •Development shifts from implementation to integration
- 14. Conclusions •Computer vision will become ubiquitous and invisible •It will be a huge creator of value, both for suppliers as well as those who leverage the technology in their applications •Deep learning will become a dominant technique (but not the only technique) •Computation distributed between the cloud and the edge •Heterogeneity in hardware becomes increasingly hidden •Development shifts from implementation to integration