System and method for use in object detection from video stream

A software product, embedded on a non-transitory computer readable medium comprising computer readable code that when executed by one or more processors, cause the processor to process input data to thereby:

• receive input data comprising at least one sequence of consecutive image frames piece of a region of interest;
• process one or more image frames for detection of one or more foreground pixels regions;
• collect adjacent foreground pixels and determining one or more blobs associated with one or more suspected foreground objects in the region of interest;
• cluster adjacent blobs forming corresponding image portions comprising clusters of said blobs;
• select one or more cluster of blobs and generate image portion indicative of one or more suspected foreground objects, and process said one or more image portions for detection of one or more foreground objects and corresponding one or more object classes;
• and
• generate output data comprising class probability for one or more objects detected in one or more frame images of said image data stream;
• wherein a number of image portions covering together all of the foreground pixel regions does not exceed a predetermined image regions threshold to thereby enable object detection with reduced computation complexity.

A system for use if object detection, the system comprising communication module, storage unit and processing utility and configured for processing input data comprising image data stream comprising at least one sequence of consecutive image frames;

• said processing utility comprises at least one processor adapted for:
• receiving one or more frames of the image data stream and processing the one or more frames to determine one or more foreground pixels regions;
• processing the one or more foreground pixels regions for collecting adjacent foreground pixels and determining one or more blobs associated with one or more suspected foreground objects;
• collecting data about the one or more blobs, clustering adjacent blobs to form corresponding image portions comprising clusters of said blobs and selecting one or more clusters of blobs and generating image portion indicative of one or more suspected foreground objects;
• and
• processing the one or more image portions for detecting objects in said image portions and determining corresponding object data comprising location and classification of one or more detected objects;
• said at least one processor thereby generates output data comprising class probability for one or more objects detected in one or more frame images of said image data stream;
• wherein said clustering adjacent blobs comprises determining a total number of image portions covering together all of the foreground pixel region, said total number does not exceed a predetermined image regions threshold.