Method and system for ground truth determination in lane departure warning

A system for ground truth determination in a lane departure warning, wherein the ground truth is determined based on a lane marking, the system comprising:

• an image capturing unit;
• an image database communicatively coupled to the image capturing unit;
• the image database configured to receive a plurality of lane images or a lane video comprising the plurality of lane images from the image capturing unit;
• an image processing unit, wherein the unit comprising:
• a processor;
• a memory coupled to the processor;
• the memory comprising:
• a receiving module for receiving the plurality of lane images or the lane video comprising the plurality of lane images as an input from the image database;
• an image processing module for
• processing the plurality of lane images received from the receiving module to obtain a plurality of time sliced images, wherein each time sliced image contains selection of predefined rows of the plurality of lane images, wherein the plurality of lane images are captured at different time frames;
• determining a histogram of each of the plurality of time sliced images and smoothening the histogram to obtain a smoothened histogram, wherein the smoothening is performed on the histogram to extract low pass filtered envelop of the histogram;
• obtaining a threshold value for each of the time sliced images by extracting a minima between a first maxima and a successive second maxima of the smoothened histogram, wherein the first maxima is a highest peak of the smoothened histogram and the successive second maxima is a following highest peak of the first maxima, wherein extracting the minima between the first maxima and the successive second maxima of the smoothened histogram is termed as Min-Between-Max-Thresholding (MBMT) technique;
• refining the threshold value of the time sliced images by:
• segmenting each of the time sliced images into a predefined number of time slices by further applying time slicing on the each time sliced image,
• calculating a statistical mean variation of each of the individual time slices;
• and
• grouping the time slices having a mean intensity within a predefined range into a cluster of time slices, wherein a predefined number of time slices having the mean intensity in the predefined range are grouped and localized and
• determining the minima between the first maxima and the successive second maxima of the histogram of the cluster of time slices by applying the MBMT technique to obtain refined threshold for each cluster of time slices;
• and
• a detection module for detecting the lane marking of the lane by extracting mid-points of the lane marking based on the refined threshold value of the time sliced images, and determining the ground truth value based on the lane marking of the lane for validating the lane departure warning based on a pixel position detected by the lane departure warning with respect to the determined ground truth.

A non-transitory computer readable medium having instructions that, when executed by a computing device cause the computing device to:

• receive a plurality of lane images a lane video comprising the plurality of lane images as an input from an image capturing unit;
• process the plurality of lane images into a plurality of time sliced images, wherein each time sliced image contains selection of predefined rows of the plurality of lane images, wherein the plurality of lane images are captured at different time frames;
• determine a histogram of each of the plurality of time sliced images and smoothening the histogram to obtain a smoothened histogram, wherein the smoothening is performed on the histogram to extract low pass filtered envelop of the histogram;
• obtain a threshold value for each of the time-sliced images by extracting a minima between a first maxima and a successive second maxima of the smoothened histogram, wherein the first maxima is a highest peak of the smoothened histogram and the successive second maxima is a following highest peak of the first maxima, wherein extracting the minima between the first maxima and the successive second maxima of the smoothened histogram is termed as Min-Between-Max-Thresholding (MBMT) technique;
• refine the threshold value of the time-sliced images by
• segmenting each of the time sliced images into a predefined number of time slices by further applying time slicing on the each time sliced image,
• calculating a statistical mean variation of the histogram of each of the individual time slices and
• grouping the individual time slices having a mean intensity within a predefined range in to a cluster of time slices, wherein a predefined number of time slices having the mean intensity in the predefined range are grouped and localized, and
• determining the minima between the first maxima and the successive second maxima of the histogram of the cluster of time slices by applying the MBMT technique to obtain refined threshold value for each cluster of time slices;
• detect the lane marking of the lane by extracting mid-points of the lane marking based on the refined threshold value of the time sliced images;
• and
• determine the ground truth based on the lane marking of the lane for validating the lane departure warning based on a pixel position detected by the lane departure warning with respect to the determined ground truth.