People detection in heavy machines applications

Abstract

In this paper we focus on improving the performance of people detection algorithm on fish-eye images in a safety system for heavy machines. Fish-eye images give the advantage of a very wide angle-of-view, which is important in the context of heavy machines. However, the distortions in fish-eye images present many difficulties for image processing. The underlying framework of the proposed detection system uses Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM). By analyzing the effect of distortions in different regions in the field-of-view and by adding artificial distortions in the training process of the binary classifier, we can obtain better detection results on fish-eye images.

Deformable parts model for people detection in heavy machines applications

“…5 left). This result conforms with the conclusion in [3]. Enriching the training dataset can handle the distortion on the people's appearance, even with the DPM approach.…”

“…Our analysis on testing the DPM approach in the context of fisheye images, begin with modifying the training database. The mix-training-dataset method [3] combined with the DPM are also taken into consideration. The dimensional-reduced-HOG is thorough evaluated in term of computational resource and performance.…”

“…The total number of positive and negative sample images in the new training dataset is the same as the original one. For more details about the process of artificially generating distorted image samples, the reader is referred to [3].…”

Multi-sensors people detection system for heavy machines

“…Moreover, construction machines often have a complicated shape and a larger size, which can also benefit from the large FOV. These features reveal the necessity to use wide angle sensors such as fisheye cameras as the main sensor for people detection and recognition [4].…”

“…This advantage is not conserved in fisheye image since the object appearance distortions depends on its distance and its angle to the camera. A quantitative analysis of this distortion phenomenon is presented in [4]. • Configuration 2: The fisheye camera is mounted at position 2 (height h = 210cm, at the forks level), looking down with an angle of 30°.…”