Recent years have witnessed significant advancement in computer vision research based on deep learning. Success of these tasks largely depends on the availability of a large amount of training samples. Labeling the training samples is an expensive process. In this paper, we present a simulated deep convolutional neural network for yield estimation. Knowing the exact number of fruits, flowers, and trees helps farmers to make better decisions on cultivation practices, plant disease prevention, and the size of harvest labor force. The current practice of yield estimation based on the manual counting of fruits or flowers by workers is a very time consuming and expensive process and it is not practical for big fields. Automatic yield estimation based on robotic agriculture provides a viable solution in this regard. Our network is trained entirely on synthetic data and tested on real data. To capture features on multiple scales, we used a modified version of the Inception-ResNet architecture. Our algorithm counts efficiently even if fruits are under shadow, occluded by foliage, branches, or if there is some degree of overlap amongst fruits. Experimental results show a 91% average test accuracy on real images and 93% on synthetic images.

Deep Count: Fruit Counting Based on Deep Simulated Learning

Most conventional autofocusing techniques based on the gradient estimator are very sensitive to noise. A new autofocusing technique which is resistive to noise generated by the CCD of video cameras is proposed. In the proposed scheme, the frequency selective weighted median (FSWM) filter is utilized to estimate the degree of focus and the fast hill-climbing search (HCS) strategy is exploited to determine the best focused image. Since the FSWM filter can not only extract high frequency components from the image, but also eliminate impulsive noise, the proposed autofocusing method employing the FSWM criterion function can estimate the degree of focus precisely. Furthermore, the proposed real-time HCS algorithm enables the video camera to continuously focus on dynamic images. Experimental results demonstrate that the proposed technique outperforms existing techniques by enhancing the accuracy of the focus value of the video camera without the influence of noise.

New autofocusing technique using the frequency selective weighted median filter for video cameras

A practical real-time auto-focus algorithm for a digital camera is presented, and it improves the reliability and speed of the auto-focus process, especially suitable for a mega-pixel high definition camera. The proposed algorithm adopts threshold gradient and edge point count technique besides focus value function, instead of the traditional two-stage climbing search algorithm that uses the focus value function only. Additionally, a relative difference ratio circuit is also proposed, which can implement adaptive step size searching to increase the searching speed. By adopting the modified algorithm on the prototype of our mega-pixel digital camera, the real-time auto-focus function is verified. The proposed algorithm is implemented in a test camera chip that has been manufactured in 0.25/spl mu/m CMOS digital process.

Modified fast climbing search auto-focus algorithm with adaptive step size searching technique for digital camera

In one embodiment, a light sensor includes four cell arrays, one for each color of the Bayer pattern, and four lenses each focusing the light coming from the scene to be captured on a respective cell array. The lenses are oriented such that at least a second green image, commonly provided by the fourth cell array, is both horizontally and vertically shifted (spaced) apart by half a pixel pitch from a first (reference) green image. In a second embodiment, the four lenses are oriented such that the red and blue images are respectively shifted (spaced) apart by half a pixel pitch from the first or reference green image, one horizontally and the other vertically, and the second green image is shifted (spaced) apart by half a pixel pitch from the reference green image both horizontally and vertically.

Compound camera sensor and related method of processing digital images

An imaging acquisition system that generates a depth map from two pictures of a three dimensional spatial scene is described. According to one aspect of the invention, the system generates the depth map based on the relative blur between the two pictures and the absolute blur contributed by the system. According to another aspect of the invention, the system calculates the depth map directly from the relative blur between the two pictures.

Depth information for auto focus using two pictures and two-dimensional Gaussian scale space theory

This paper describes an implementation method for the people counting system which detects and tracks moving people using a fixed single camera. This system counts the number of moving objects (people) entering the security door. Moreover, the detected objects are tracked by the proposed tracking algorithm before entering the door. The proposed system with Intel Pentium IV operates at an average rate of 10 frames a second on real world scenes where up to 6 persons come into the view of a vertically mounted camera.

/pdf/real-time-vision-based-people-counting-system-for-the-4f2ocpdd55.pdf

Real-time Vision-based People Counting System for the Security Door

The regularising parameter of the energy function in the graph-cut based image segmentation methods should be carefully determined since it strongly affects the image segmentation result. Presented is a modified graph-cut technique using a novel energy function without the regularising parameter for image segmentation. In addition, a mixture colour model is employed to apply the proposed technique to colour image segmentation effectively. Experimental results show that the proposed method outperforms the conventional graph-cut algorithm in terms of segmentation accuracy and ease of use.

Image segmentation based on modified graph-cut algorithm

This paper introduces a novel nonlinear filtering structure: the linear combination of weighted medians (LCWM). The proposed filtering scheme is modeled on the structure and design procedure of the linear-phase FIR highpass (HP) filter in that the linear-phase FIR HP filter can be obtained by changing the sign of the filter coefficients of the FIR lowpass (LP) filter in the odd positions. The HP filter can be represented as the difference between two LP subfilters that have all positive coefficients. This representation of the FIR HP filter is analogous to the difference of estimates (DoE) such as the difference of medians (DoM). The DoM is essentially a nonlinear HP filter that is commonly used in edge detection. Based on this observation, we introduce a class of LCWM filters whose output is given by a linear combination of weighted medians of the input sequence. We propose a method of designing the 1-D and 2-D LCWM filters satisfying required frequency specifications. The proposed method adopts a transformation from the FIR filter to the LCWM filter. We show that the proposed LCWM filter can offer various frequency filtering characteristics including "LP," "bandpass (BP)," and "HP" responses.

Design of linear combination of weighted medians

In this paper, we propose a real-time people counting system with a single camera for security inside the building. The camera is hung from the ceiling of the door so that the image data of the passing people are not fully overlapped. The implemented system recognizes people movement along various directions. To track people even when their images are partially overlapped, the proposed system analyzes the captured image and estimates important feature information. Then each bounding box enclosing each person in the tracking region are tracked. The convex hull of each individual obtained by the proposed algorithm provides more accurate tracking information. Through extensive tests in various environments, the proposed system exhibits the correct people counting results of over 96%.

Kang-Sun Choi

Papers

New autofocusing technique using the frequency selective weighted median filter for video cameras

Real-time Vision-based People Counting System for the Security Door

Image segmentation based on modified graph-cut algorithm

Design of linear combination of weighted medians

Real-time system for counting the number of passing people using a single camera