Karol Piniarski

Pedestrian detection in low resolution night vision images

This paper presents a test of pedestrian detection in low resolution night vision infrared images. An image feature extractor based on histograms of oriented gradients followed by a Support Vector Machine (SVM) classifier are evaluated, optimized and used. Tests performed on three different night vision infrared datasets show that the classification quality of the proposed method is very high even in very low resolutions of images. In practice, large frame size for analysis not always improves the classification effectiveness, but always requires more time for processing.

Mobile applications for driver and pedestrian assistance

In this paper two applications, which fit into the current trend of the development of mobile devices, are described. The first is designed for the driver and the second for the pedestrian assistance. Two traffic events like traffic lights change from the red to the green and vehicle approaching to pedestrian are detected. Proper warnings are generated and the suitable reaction is performed. The car may start movement without unnecessary delay and the pedestrian may change the movement trajectory in order to avoid an accident. Thus, the proper traffic flow and urban safety improvement are supported. Results of initial experiments show a high efficiency of proposed mobile solutions.

Multi-branch classifiers for pedestrian detection from infrared night and day images

This paper presents a modified multi-branch classifier of pedestrians from far infrared (FIR) night and day images. The solution is accurate, fast, and especially best suited for all realtime applications where pedestrians may appear in many distances to the camera, like in cars and CCTV. Two proposed training methods of the classifier, namely full-scale and partial-scale training were deeply tested. Results show increased efficiency of the classification process (by up to 3 %) with similar processing time in comparison to a single classifier. All tests were conducted using Adaboost classifier, but generally, the results should be consistent for other classifiers.

Efficient pedestrian detection with enhanced object segmentation in far IR night vision

This paper presents a pedestrian detection system with enhanced object segmentation procedure working on a far infrared (FIR) video. To make the object detection more accurate on the FIR images, we propose an enhanced segmentation procedure with two thresholds and the region enlargement. This combination allowed a significant reduction of the region of interests (ROIs) for further processing. Experiments performed on demanding public dataset show a significant increase of the pedestrian detection performance (up to 33 frames per second) with the accuracy comparable with state-of-the-art algorithms.

HDR tonal mapping algorithm for mobile devices

In this paper the application for generation of HDR image based on two consecutive images (underexposed and overexposed) for Android mobile operating system is presented. The implemented software preserves a lot of image details and maintains a low execution time. These features are particularly important for pictures taken using mobile devices in emergency situations. Such photos may constitute evidence that a threat occurred, was properly recognized, or someone committed a crime. HDR images can be also used in mobile systems for supporting pedestrians or drivers. Obtained results indicate on a high effectiveness of the presented solution.