Real-Time Motion Detection and Surveillance using Approximation of Image Pre-processing Algorithms

A typical computer vision task for surveillance may include the following step; Image Acquisition, Image Pre-processing, Image Enhancement, Image Segmentation, Feature Extraction and Identification of object. Each step in the pipeline is critical since poor performance at any stage may affect the overall result of a computer vision application. Among the steps indicated, the image pre-processing step is crucial for the performance of a model. However, the time taken by the pre-processing algorithms to complete their tasks generally contributes to the overall running time of a computer vision application. Various techniques and algorithms have been proposed to help reduce the time complexities associated with image pre-processing algorithms. Among the many algorithms are the Adaptive Approximated Median Filtering Algorithm for Impulse Noise Reduction, A Robust Median-based Background Updating Algorithm and Fast Generation of Image's Histogram Using Approximation technique for Image Processing Algorithms. These proposed methods used approximation of existing techniques to significantly reduce the time complexity associated with the algorithms. However, their resultant effects in the pipeline of a typical computer vision application had not been evaluated. This paper therefore evaluated the effectiveness of the algorithms on motion detection and movement tracking computer vision application. The experimental results indicate that these pre-processing algorithms can significantly reduce the overall running times for motion detection and scene boundary monitoring, and therefore can easily replace original ones for real-time motion and object tracking.