In today’s world, embedded systems have become an integral part of our daily lives. The advancements in computer vision and machine learning, in conjunction with the abundance of computing power, have made it possible for these systems to perform other tasks such as image processing and video analytics; giving rise to embedded vision systems. However, the computer vision algorithms cannot always address the computational costs required due to the complexity and weight of the embedded computer processing architecture. Therefore, the execution performance of the processing algorithms should be investigated further. The basic background subtraction algorithm is found to have low complexity, but it may affect the segmentation quality of the foreground result. In this thesis, a new intelligent computer vision processing-based technique is proposed to detect motion (vehicle) in an embedded travel time estimation. Initially, a light, fast, and efficient background subtraction is needed for this system to extract the subject from frame sequences. A new frame difference-based method was proposed with a background counter map to conduct background determination decisions. Two different versions of the algorithm, which are proposed BGS 1 and 2, were developed for different purpose (noisy and multimodal background) and evaluated using a quantitative segmentation quality assessment method based on relevant parameters. Moreover, the algorithms were also analyze the CPU and memory performance and execution time before the final result was used and applied on the embedded-based travel time estimation. Two algorithms were evaluated using a PC to determine the best parameter value for the algorithm application. The findings proved that the proposed methods performed excellently in terms of segmentation quality. Proposed BGS 2 performed better for every measurement (F-measured—31.67% improvement, SSIM—11.84% improvement, 9.78 % less error, and FSD—11.02 % improvement) compared to proposed BGS 1. Based on this value, both algorithm is implement on the Raspberry Pi and it used 25%-26% of CPU and 4.5%-4.7% of memory. But both algorithm maintained the segmentation quality score as on PC and provided a similar segmentation quality score compare to PBAS and MOG2 with limitation to the provided dataset. Both also execute fast as the frame difference algorithm. Based on these results, proposed BGS was implemented for travel time estimation and analyzed. Travel time estimation worked well by consumed 41%-56% of CPU with 4.5% - 5.5% of memory.
