Performance analysis of CPU & GPU for Real Time Image/Video

Computer vision algorithms are used in applications which require the given system to process and display images or video inputs. However, this will be computationally intensive on the machine that is performing the automated task based on the visual inputs and results in a large overhead or a lag between input and output processed video. To address this issues, parallel processing is a very useful strategy and can be achieved by dividing the computational tasks among the given hardware of the system which would make use of it efficiently and would work around the limitations of the hardware in the system. In this paper, the parallelism is achieved by multi-threading the algorithm that will divide the computation among the number of cores in the CPU and further optimize by dividing the load with both CPU and GPU for an efficient use of the system hardware and to obtain an optimized result. When the algorithms are executed without any optimization, the obtained output video fps is high and is undesirable. In this proposed method, a speedup factor of 91 times is recorded in AIELI algorithm.