Artificial Intelligence/Machine Learning, ASA(ALT), Direct to Phase II
Robust Computer Vision for Better Object Detection with Limited Training Data
Selectees
Objective
The U.S. Army seeks to experiment with innovative artificial intelligence and machine learning approaches for object identification and imagery scene analysis.
Description
With the increasing availability of digital imagery, including satellite data for electro-optical/infrared, synthetic aperture radar and full-motion video, there is a growing need for automated methods to efficiently process and analyze vast amounts of multi-modal data.
One critical application is the identification of objects of interest within imagery data or the scene generated by the imagery. This can provide valuable insights and facilitate decision-making processes in various fields such as military intelligence, environmental monitoring, transportation management and security surveillance.
Phase I
The Army will only accept Direct to Phase II proposals for contracts worth up to $2,000,000 over an 18-month performance period.
Proposers interested in submitting a DP2 proposal must provide documentation to substantiate that the technology meets the scientific, technical merit and feasibility equivalent to a Phase I project. Documentation can include data, reports, specific measurements and the success criteria of a prototype.
This SBIR solicitation will explore robust AI/ML object detection techniques for computer vision that do not rely on the extensive availability of labeled training data. Foundational knowledge and methods already exist, making feasibility studies unnecessary.
Computer vision algorithms using handcrafted mathematical features, which include edge detection and scale-invariant feature transform, are still effective for certain tasks while offering faster run times.
Evolutionary algorithms, such as Neuroevolutionary of Augmenting Topologies, can optimize the parameters of a computer vision system and combine with other methods such as handcrafted features and various neural networks architectures. These help to form hybrid approaches with less dependence on extensive training data.
Newer techniques based on transformers, and referred to as foundational models, have shown extraordinary ability to generalize to new tasks without requiring use case specific training data. All these computer vision technologies can function within academic and industrial settings, even reaching sufficient maturity for deployment in commercial products such as level-two self-driving cars and vision language models like Google’s Gemini.
Vendors can leverage these foundational technologies for the SBIR solicitation and adapt them for Department of Defense and Army use cases without requiring a feasibility study.
Phase II
During the DP2, firms should develop and implement novel or hybrid AI/ML models for object detection that do not rely on extensive training data. Vendors should also develop training models in Project Linchpin’s AI Unclassified Operations Environment using Linchpin data for DoD use cases.
Phase III
- Autonomy: Detecting objects and obstacles for self-driving cars, robots and drone delivery initiatives.
- Retail: Analyze shopping behavior in store to gain insights into product interactions and contactless checkout.
- Public safety: Detection of unauthorized objects or individuals in manufacturing, logistics and construction sectors.
- Traffic management: Monitor roads to optimize traffic flow and reduce congestion.
- Enhanced security: Improving security systems for access control and surveillance purposes.
- Agriculture: Computer vision can help prediction and plant monitoring to detect diseases.
Computer vision solutions in the private sector encompass a wide range of applications, from object detection and recognition to healthcare and agriculture. Companies such as Amazon, Google and Microsoft offer cloud-based object detection and recognition services. Meanwhile Face++, Kairos and NEC provide facial recognition solutions. Additionally, companies like IBM, Cisco, and Huawei offer video analytics solutions while ABB, Kuka, and FANUC provide vision-guided robotics and automation solutions.
Submission Information
All eligible businesses must submit proposals by noon, ET.
To view the full solicitation details, click here.
For more information, and to submit your full proposal package, visit the DSIP Portal.
Applied SBIR Help Desk: usarmy.pentagon.hqda-asa-alt.mbx.army-applied-sbir-program@army.mil
References:
- B. Amjoud and M. Amrouch, “Object Detection Using Deep Learning, CNNs and Vision Transformers: A Review,” in IEEE Access, vol. 11, pp. 35479-35516, 2023, doi: 10.1109/ACCESS.2023.3266093.
- L. Jiao et al., “New Generation Deep Learning for Video Object Detection: A Survey,” in IEEE Transactions on Neural Networks and Learning Systems, vol. 33, no. 8, pp. 3195-3215, Aug. 2022, doi: 10.1109/TNNLS.2021.3053249.
- Y. Bi, B. Xue, P. Mesejo, S. Cagnoni and M. Zhang, “A Survey on Evolutionary Computation for Computer Vision and Image Analysis: Past, Present, and Future Trends,” in IEEE Transactions on Evolutionary Computation, vol. 27, no. 1.
- Digital Imagery; Objects of Interest; Sensor Data; AI/ML; Scale-Invariant Feature Transform; Neuroevolutionary of Augmenting Topologies
Selectees
Objective
The U.S. Army seeks to experiment with innovative artificial intelligence and machine learning approaches for object identification and imagery scene analysis.
Description
With the increasing availability of digital imagery, including satellite data for electro-optical/infrared, synthetic aperture radar and full-motion video, there is a growing need for automated methods to efficiently process and analyze vast amounts of multi-modal data.
One critical application is the identification of objects of interest within imagery data or the scene generated by the imagery. This can provide valuable insights and facilitate decision-making processes in various fields such as military intelligence, environmental monitoring, transportation management and security surveillance.
Phase I
The Army will only accept Direct to Phase II proposals for contracts worth up to $2,000,000 over an 18-month performance period.
Proposers interested in submitting a DP2 proposal must provide documentation to substantiate that the technology meets the scientific, technical merit and feasibility equivalent to a Phase I project. Documentation can include data, reports, specific measurements and the success criteria of a prototype.
This SBIR solicitation will explore robust AI/ML object detection techniques for computer vision that do not rely on the extensive availability of labeled training data. Foundational knowledge and methods already exist, making feasibility studies unnecessary.
Computer vision algorithms using handcrafted mathematical features, which include edge detection and scale-invariant feature transform, are still effective for certain tasks while offering faster run times.
Evolutionary algorithms, such as Neuroevolutionary of Augmenting Topologies, can optimize the parameters of a computer vision system and combine with other methods such as handcrafted features and various neural networks architectures. These help to form hybrid approaches with less dependence on extensive training data.
Newer techniques based on transformers, and referred to as foundational models, have shown extraordinary ability to generalize to new tasks without requiring use case specific training data. All these computer vision technologies can function within academic and industrial settings, even reaching sufficient maturity for deployment in commercial products such as level-two self-driving cars and vision language models like Google’s Gemini.
Vendors can leverage these foundational technologies for the SBIR solicitation and adapt them for Department of Defense and Army use cases without requiring a feasibility study.
Phase II
During the DP2, firms should develop and implement novel or hybrid AI/ML models for object detection that do not rely on extensive training data. Vendors should also develop training models in Project Linchpin’s AI Unclassified Operations Environment using Linchpin data for DoD use cases.
Phase III
- Autonomy: Detecting objects and obstacles for self-driving cars, robots and drone delivery initiatives.
- Retail: Analyze shopping behavior in store to gain insights into product interactions and contactless checkout.
- Public safety: Detection of unauthorized objects or individuals in manufacturing, logistics and construction sectors.
- Traffic management: Monitor roads to optimize traffic flow and reduce congestion.
- Enhanced security: Improving security systems for access control and surveillance purposes.
- Agriculture: Computer vision can help prediction and plant monitoring to detect diseases.
Computer vision solutions in the private sector encompass a wide range of applications, from object detection and recognition to healthcare and agriculture. Companies such as Amazon, Google and Microsoft offer cloud-based object detection and recognition services. Meanwhile Face++, Kairos and NEC provide facial recognition solutions. Additionally, companies like IBM, Cisco, and Huawei offer video analytics solutions while ABB, Kuka, and FANUC provide vision-guided robotics and automation solutions.
Submission Information
All eligible businesses must submit proposals by noon, ET.
To view the full solicitation details, click here.
For more information, and to submit your full proposal package, visit the DSIP Portal.
Applied SBIR Help Desk: usarmy.pentagon.hqda-asa-alt.mbx.army-applied-sbir-program@army.mil
References:
- B. Amjoud and M. Amrouch, “Object Detection Using Deep Learning, CNNs and Vision Transformers: A Review,” in IEEE Access, vol. 11, pp. 35479-35516, 2023, doi: 10.1109/ACCESS.2023.3266093.
- L. Jiao et al., “New Generation Deep Learning for Video Object Detection: A Survey,” in IEEE Transactions on Neural Networks and Learning Systems, vol. 33, no. 8, pp. 3195-3215, Aug. 2022, doi: 10.1109/TNNLS.2021.3053249.
- Y. Bi, B. Xue, P. Mesejo, S. Cagnoni and M. Zhang, “A Survey on Evolutionary Computation for Computer Vision and Image Analysis: Past, Present, and Future Trends,” in IEEE Transactions on Evolutionary Computation, vol. 27, no. 1.
- Digital Imagery; Objects of Interest; Sensor Data; AI/ML; Scale-Invariant Feature Transform; Neuroevolutionary of Augmenting Topologies