1374484 - Crowd Density Detection

Pedestrians are an important and most vulnerable component of transportation communities. Right now, vehicle radars, cameras, and additional sensors are invented and widely used to gain intel on the surrounding environment. Most sensors today such as radar or metal object detection strive to detect vehicles, not pedestrians. The research to detect pedestrians morphed into real-time pedestrian detection in real traffic application scenarios. It is flexible, cost-effective, and capable of detecting tiny pedestrian pictures and silhouettes. By using techniques to enhance the current accuracy and information there is on pedestrians, proposed novel detectors such as IENet make full use of the internal and external information which leads to better detection. Utilizing an Inception Parallel Attention Network (IPAN) will extract more information from small objects and take intel from surrounding contextual correlations. YOLO is improved with a new loss function to reduce the impact of the imbalance problems and will adapt to fit multi-scale objects. Object detection algorithms are improved through the use of SVM, median filter, and achieving higher accuracy. Research has found the technique of cropping crowd regions to be exceptionally useful by first generating an attention map to generate a density map and then determining the high-density regions and clustering them. So, the research on multi-column techniques, and dilated filters, will be beneficial for multi-scale counting to further solve inconsistencies in density.