Design of an embedded smart camera for person detection and identification

Current video surveillance systems transport huge amounts of data. Smart cameras are developed to provide cameras with the capability to locally analyze video therefore considerably reducing the amount of data to be transported. To fulfil their objectives, smart cameras must provide high performance computing to handle complex image processing algorithms. They must also be designed to meet cost, size, weight, power and energy constraints. Moreover, to achieve higher degrees of operational capabilities, embedded platform flexibility and easy programmability should be maintained. This project focuses on the design of embedded smart camera architectures targeting person detection and identification systems. More specifically, All Programmable System on Chip (APSoC) platform solutions are selected for system prototyping. The support of an appropriately selected hardware and software APSoC platform for fast prototype development is going to be demonstrated through the design of a particular application of person detection and identification. The performance of such vision application solely depends on the selected image feature descriptors. An ideal feature descriptor must be robust, discriminative, and fast enough to compute. This project motivates both COVariance (COV) and MultiScale COVariance (MSCOV) descriptors. Improved versions of such descriptors featuring high performance accuracies at reduced processing power and memory requirements are proposed. Prototype development and performance analysis of a smart camera architecture implementing the proposed new descriptor are evaluated for person detection and identification. The successful execution of this project should produce 1 to 2 conference papers, and 1 journal papers

Current video surveillance systems transport huge amounts of data. Smart cameras are developed to provide cameras with the capability to locally analyze video therefore considerably reducing the amount of data to be transported. To fulfil their objectives, smart cameras must provide high performance computing to handle complex image processing algorithms. They must also be designed to meet cost, size, weight, power and energy constraints. Moreover, to achieve higher degrees of operational capabilities, embedded platform flexibility and easy programmability should be maintained. This project focuses on the design of embedded smart camera architectures targeting person detection and identification systems. More specifically, All Programmable System on Chip (APSoC) platform solutions are selected for system prototyping. The support of an appropriately selected hardware and software APSoC platform for fast prototype development is going to be demonstrated through the design of a particular application of person detection and identification. The performance of such vision application solely depends on the selected image feature descriptors. An ideal feature descriptor must be robust, discriminative, and fast enough to compute. This project motivates both COVariance (COV) and MultiScale COVariance (MSCOV) descriptors. Improved versions of such descriptors featuring high performance accuracies at reduced processing power and memory requirements are proposed. Prototype development and performance analysis of a smart camera architecture implementing the proposed new descriptor are evaluated for person detection and identification. The successful execution of this project should produce 1 to 2 conference papers, and 1 journal papers